Рубрика: AI

Abu Dhabi-based AI group G42 on Tuesday announced the release of NANDA 87B, an open-source Hindi-English large language model with 87 billion parameters, an upgrade to its earlier NANDA model.

NANDA 87B is accessible as an open-weight resource on the MBZUAI Hugging Face page, allowing creators, developers, and businesses to utilise and expand its features.

The model has been developed by Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) in collaboration with Inception, a G42 company, and chipmaker Cerebras.

Built on Llama-3.1 70B, NANDA 87B has been trained on more than 65 billion Hindi tokens using a Hindi-centric tokeniser to improve efficiency in training and inference.

“India deserves world-class technology that speaks its language. NANDA 87B is a major step in that direction,” said Manu Jain, chief executive of G42 India, adding that the model is intended to support innovation across education, entertainment and enterprise use cases in India’s AI ecosystem.

G42 said the model is designed to handle formal Hindi, casual speech and Hinglish, and performs tasks such as translation, summarisation, instruction following and transliteration. The company added that safety and cultural alignment were part of the model’s design to ensure responsible outputs.

Richard Morton, executive director at MBZUAI’s Institute of Foundation Models, said the release marked progress in expanding access to advanced language technology. “NANDA marks an important milestone in bringing high-quality, open-access language technology to one of the world’s largest linguistic communities,” he said

The model was trained on Condor Galaxy, an AI supercomputing system built by G42 and Cerebras.

The post Abu Dhabi-based G42 Unveils Open-Source Hindi-English Model NANDA 87B appeared first on Analytics India Magazine.

The world is in the middle of an AI compute crunch. Demand for GPUs has exploded far beyond the pace at which traditional cloud providers can build, provision, or price capacity. As startups scramble, enterprises stall projects, and researchers wait in line for hardware, a new model for scaling compute is emerging, one that aggregates unused or underutilised GPU supply and delivers it as a flexible, transparent marketplace.

The global race for AI compute has pushed enterprises, researchers and startups to look beyond traditional hyperscalers. In an interaction with AIM, io.net CEO Gaurav Sharma said he believes decentralised GPU networks may be the only model that can scale fast enough, and India may end up playing a far larger role in this sector than anyone expects.

Sharma’s journey spans two decades across Linux kernel engineering, AWS, Agoda, eBay, and Binance. In recent years, he noticed the shortage of compute in the AI space. That bottleneck, he said, was directly hurting builders. The answer was to create a platform that could return control, affordability and transparency to developers.

Decentralisation, But Not As India Knows It

Sharma argued that decentralisation is widely misunderstood, especially in markets where centralised cloud still dominates.

“When people think of decentralisation, they’re not very clear what we are talking about… decentralisation can mean different things to different people,” he said.

io.net’s network pools GPUs from individual users, data centres and global contributors, then routes workloads based on availability, stability and price. The pitch to customers is not ideological. “We don’t even talk in terms of decentralisation,” Sharma noted. Companies simply care about reliability and cost, not Web3 philosophy.

He compared the model to MakeMyTrip for GPUs, “aggregating supply from multiple suppliers… whatever they need, we give them.”

In his view, cloud incumbents have little incentive to fix today’s pricing and capacity barriers. A decentralised marketplace can fill that gap faster than building new data centres, which take months just to procure hardware.

India is not yet a decentralisation-first market, but paradoxically, that makes it more important to io.net’s plan.

Sharma highlighted three key advantages for training AI models in India, especially for projects prioritising speed and cost. First, a deep reservoir of technical talent with necessary GPU configuration expertise, often scarce in the US, second, cost-efficient operations due to favourable labour and energy economics, lowering GPU running costs compared to Western markets.

And the third: faster scaling capabilities, as India’s large engineering talent pool enables rapid provisioning of global AI workloads.

Making the Marketplace Work, Tackling Challenges

io.net is tackling the immediate AI compute shortage by adopting a rapid Web3-based scaling strategy, bypassing the slow, multi-year expansion model typical of traditional networks.

Instead of relying on a large conventional funding round to incentivise data centre integration, the company used a $40 million Web3 raise and tokenomics. This approach served the dual purpose of quickly building the necessary network infrastructure to overcome the “cold start problem” and fostering community engagement through airdrops and continuous product testing with minimal capital expenditure.

However, this accelerated scaling introduces two unavoidable operational challenges. The first is the continuous maintenance of quality and data accuracy, as platforms are inherently vulnerable to degradation over time. Like any marketplace, io.net must consistently refine its network with new data to ensure the reliability of its claimed GPU and data centre inventory, understanding that occasional negative customer experiences are a business reality requiring patience.

The second challenge is overcoming customer scepticism. Even with a strong inventory, new users are typically cautious, starting with small-scale testing (e.g., 10-15 GPUs) over a period of two to three months before committing to greater demands. While this initial verification phase slows down potential exponential growth, it is considered a necessary element for building “flywheel momentum.” This caution is expected to diminish as io.net’s reputation solidifies through successful adoption by peer networks and established companies.

io.net earns through a platform fee and revenue share with data centres. Operating costs scale minimally because the platform is horizontally scalable.

Who’s Using io.net Today?

Sharma listed a wide mix of customers — from IIT Bombay and UC Berkeley to Eros Now and robotics platform Frodobots. Startups working on audio generation, image models and voice-to-song synthesis use the GPUs heavily.
Through partnerships with Antler and YC, around 15–20, or more early-stage companies now rely on io.net for compute.

He highlighted AI startup Wondera.ai, which uses io.net for an LLM that can generate songs in the voice of specified artists.

In just six months of monetisation, io.net crossed $25 million in revenue, with larger contracts in the pipeline. The company expects India to become a major contributor to supply, engineering and customer demand.

Sharma believes hyperscalers cannot build data centres fast enough to keep pace with the explosion in AI workloads. A decentralised GPU marketplace, he said, may be the only model that can scale at internet speed.

The post From Shortages to Scale, io.net’s Approach to Rewriting AI Compute Access appeared first on Analytics India Magazine.

ADaSci today announced the launch of the Certified Forward Deployed Engineer (CFDE) program, a 30-hour, self-paced certification built for engineers who work at the intersection of customers, engineering, and production systems. As AI-first products, cloud platforms, and enterprise software continue to converge, the role of the Forward Deployed Engineer has moved from niche to essential. CFDE is designed to formalize that role with a globally recognised, skills-first certification.

This program is aimed at professionals who do not just build software, but ship outcomes. Engineers who translate vague business problems into concrete architectures. Technologists who prototype fast, deploy securely, and stay accountable after go-live. CFDE is built for the realities of modern enterprise delivery, not textbook scenarios.

Why Forward Deployed Engineering Matters Now

Enterprises today face a paradox. Technology stacks are more powerful than ever, yet adoption often fails at the last mile. Products are capable, but customers struggle to operationalise them. This gap is where Forward Deployed Engineers operate.

Forward Deployed Engineers sit close to customers while retaining deep engineering ownership. They are responsible for solution design, rapid prototyping, production deployment, and long-term reliability. As AI systems become agentic, data-driven, and deeply embedded in workflows, this role has become critical across SaaS, AI platforms, cloud providers, and enterprise IT teams.

The CFDE program is built to reflect this shift and prepare engineers for roles that demand speed, judgement, and technical depth.

Call to action
Explore the CFDE program and understand the role in detail at https://adasci.org

What the CFDE Program Delivers

The ADaSci Certified Forward Deployed Engineer program is a comprehensive, hands-on learning experience spanning 30 hours of structured content. It combines engineering fundamentals with real enterprise execution.

Participants learn how to move from ambiguous customer requirements to precise system architectures. The program covers rapid solutioning, production-grade engineering practices, and post-deployment support in complex environments. Every module is designed to translate directly to field execution rather than abstract learning.

Through guided labs, real-world case studies, and end-to-end solution-building exercises, participants gain confidence in delivering customer-facing systems that work reliably under real constraints.

Call to action
View the full curriculum and learning outcomes at https://adasci.org

Built for Enterprise Reality

One of the defining aspects of CFDE is its focus on enterprise readiness. Modern deployments are not just about code that runs. They require secure APIs, observability, compliance awareness, and operational discipline.

The program introduces participants to production-grade concerns such as SLA design, OpenTelemetry-based tracing, security best practices, and compliance frameworks including SOC2, ISO, and HIPAA. These are skills often learned too late, on the job, under pressure. CFDE brings them forward into structured learning.

By the end of the program, participants can demonstrate not just functional builds, but systems designed for scale, reliability, and accountability.

Call to action
Understand how CFDE prepares you for production environments at https://adasci.org

A Full-Stack View of Modern Engineering

CFDE is intentionally broad. Forward Deployed Engineers are expected to work across boundaries, not within silos. The curriculum spans APIs, microservices, cloud platforms, data pipelines, AI systems, and DevOps workflows.

Participants gain exposure to agentic AI architectures, retrieval-augmented generation pipelines, modern data engineering, CI CD workflows, containerisation, and orchestration using Docker and Kubernetes. The goal is not mastery of every tool, but the ability to design, deploy, and troubleshoot end-to-end systems independently.

This breadth positions CFDE holders as all-round engineers who can operate effectively across teams and stakeholders.

Call to action
See how CFDE builds end-to-end capability at https://adasci.org

Designed for Career Acceleration

Forward Deployed Engineers often sit at the centre of high-impact projects. They work directly with customers, influence adoption, and are trusted with complex deployments. As a result, these roles often come with higher visibility, faster growth, and greater responsibility.

The CFDE certification validates readiness for such positions. It signals to employers that the holder can handle ambiguity, move fast without breaking systems, and represent engineering teams in customer-facing environments.

For software engineers, solutions engineers, technical consultants, and FDE aspirants, CFDE provides a clear path to differentiated career growth.

Call to action
Position yourself for high-impact roles with CFDE at https://adasci.org

Flexible and Self-Paced by Design

The CFDE program is fully self-paced and structured to fit around professional commitments. The 30-hour curriculum can be completed at a learner’s own speed, with the certification exam available to schedule throughout the year.

This flexibility allows working professionals to upskill without stepping away from active roles, while still earning a rigorous, globally recognised credential backed by ADaSci.

Call to action
Start learning on your schedule at https://adasci.org

Who Should Enrol

The CFDE program is designed for software engineers, solutions engineers, customer-facing architects, technical consultants, and professionals aspiring to Forward Deployed Engineer roles. It is particularly relevant for those working in AI-first companies, cloud-native environments, enterprise SaaS, and digital transformation teams.

If your role requires you to bridge customers and engineering, CFDE is built for you.

Call to action
Check eligibility and enrol today at https://adasci.org

About ADaSci

ADaSci, the Association of Data Scientists, is a global professional body dedicated to advancing skills, standards, and careers across data science, AI, and modern engineering disciplines. Its certifications are recognised internationally and aligned with real industry needs.

The launch of the Certified Forward Deployed Engineer program reflects ADaSci’s commitment to staying ahead of how technology roles are evolving in practice, not just in theory.

Call to action
Learn more about ADaSci and its certifications at https://adasci.org

The post ADaSci Launches the Certified Forward Deployed Engineer Program appeared first on Analytics India Magazine.

For decades, tax research has followed a stubbornly familiar rhythm. A complex question lands on a consultant’s desk. They open case law databases, dig through circulars, scan judgments, draft a view, send it up the chain, wait for reviews, revise, and finally respond. Roughly 10-20 hours later, a carefully worded answer goes out.

The process worked, but it never really changed. That lag is what Deloitte decided to challenge.

Earlier this month, Deloitte India launched Tax Pragya, an agentic AI platform for tax research and summarisation that aims to do something radical in a conservative profession. It compresses hours of deep technical tax research into minutes, without compromising trust in accuracy, sources or judgement.

Trained on more than 1.2 million tax cases and over 5,000 Deloitte technical papers, the platform spans income tax, GST and transfer pricing on a single interface.

Building it was no small task. The team spent 18 months building it, following several rounds of internal tests and experiments with hundreds of Deloitte’s own expert employees.

Sumit Singhania, partner and chief strategy and digital transformation officer, tax at Deloitte South Asia, told AIM that the idea essentially stemmed from internal conversations he had with teams and many of Deloitte’s clients. It was revealed that deep technical tax research is still being done the same way as 10 years ago.

“In the time that we live today with the advancement of technology, we cannot spend the same amount of hours doing technical research,” he said. The ambition was not incremental improvement. Deloitte wanted to compress research time from hours into minutes without compromising accuracy.

That constraint shaped every design choice.

Under the hood, Tax Pragya runs on a private LLM enriched with a massive, tightly curated knowledge base. Public jurisprudence data from over a million cases is fused with more than 5,000 Deloitte-authored technical papers and internal knowledge assets.

“The intersection of the public database and the Deloitte knowledge base brings a very powerful source for research,” Singhania said. “By adding the power of generative AI to this, the speed has been made such a fast and agile outcome.”

On a similar note, EY India, in collaboration with Taxmann, launched Taxmann.AI, an AI-powered platform tailored for tax and legal professionals, which also offers an AI bot for responses. PwC also features Navigate Tax hub, a generative-AI-based tax colleague, aimed at automating tax work for firms.

Accuracy, Not Novelty, was the Real Challenge

Deloitte aims to be different. Debashish Banerjee, partner and data science and applied AI leader at the company, said that the foundational problem has always been hallucination.

In tax, fabricated answers are not merely inconvenient. They are unacceptable.

“Whenever clients would use any open source LLM or any other models, there’s a big risk of hallucination, and especially for businesses like tax that we are trying to solve for, it’s like a complete no-no,” Banerjee told AIM. “It’s almost like those aircraft engines where you need to run at a 99.99% accuracy.”

The solution started with data discipline. Every judicial precedent, every law, every internal paper was vetted by functional experts before entering the system.

Model selection came later. Deloitte tested multiple foundational models across hyperscalers through a blinded validation process. This included Microsoft’s Omni, AWS Lambda, Llama, Mistral and Gemini.

Tax experts evaluated responses without knowing which model produced them. The company then landed on Microsoft OmniParser 4.0 as the best model.

That choice led Deloitte to build on Microsoft Azure, though Banerjee stressed the setup remains flexible as models evolve. The firm has also invested in on-prem GPU infrastructure to experiment with smaller proprietary models in the future.

Refusing to Answer When it Doesn’t Know

What sets Tax Pragya apart is not just what it answers, but what it refuses to answer.

“To the extent, if our model does not know an answer, it will come back and say ‘Sorry, I don’t know the answer,’” Banerjee said. He contrasted this with open-source LLMs, which try to generate an answer irrespective of the question.

That restraint was tested internally for nearly a year. Deloitte treated itself as client zero. “We don’t want to launch something that we haven’t tested on ourselves,” Singhania said. “We gave it into the hands of 4,000 of our tax professionals.”

For 10 months, the platform was stress-tested against the most complex real-world queries Deloitte receives. A dedicated centre of excellence tracked feedback, accuracy and failure cases in production conditions.

The numbers convinced them. Roughly 60% of Deloitte’s tax professionals now use the tool actively. Hundreds of users spend more than 15 minutes on it every week. Hallucination rates dropped to near zero.

This internal confidence paved the way for a public launch.

Unlike conventional tools that merely retrieve documents, Tax Pragya handles contextual queries, performs background research and delivers substantiated answers. An interactive chatbot allows users to explore case law and migrate documents into conversations with a single click.

Becoming AI-First

“Tax Pragya represents a pivotal shift in how tax teams of businesses, large or small, can use responsible AI to access technical insights at speed,” Gokul Chaudhri, president of tax, Deloitte India, said in a statement.

The rollout itself reflects Deloitte’s cautious ambition. The first wave targets 500 clients through closed-door sessions. A second wave in January expands access to 5,000 clients.

Singhania frames this less as a revenue play and more as an exercise of being AI-first.

“If you don’t do this, we will be left behind,” he said. “This has become table stakes.” That view extends beyond client delivery into talent strategy. Rule-based work, he argues, no longer attracts or retains skilled professionals. “We want to be a tech-first, tech-savvy business.”

Banerjee echoed that pragmatism. Deloitte’s broader agentic AI strategy is built around solving specific business problems, not chasing hype.

That philosophy was sharpened by industry missteps, including recent public cases of AI-generated fake citations; it agreed to issue a $440,000 refund to the Australian government for an AI-assisted report filled with fabricated references and hallucinated citations, which was published in July.

Singhania did not dismiss those incidents. For Deloitte, he says the lesson reinforced the importance of guardrails, curated data and models trained to say no.

Looking ahead, Singhania sees Tax Pragya as the beginning, not the end. AI will increasingly layer itself across compliance tools, advisory workflows and decision systems. Over time, Deloitte wants to take this capability beyond large enterprises.

“Our vision also is that this platform should be the MSMEs deep into the Bharat markets,” he said. “At a reasonably affordable price point.”

For Banerjee, the most significant risk is not technology failure but adoption.

“The tax profession is already being disrupted,” he said. “People will have more time to do more analytical and critical strategic work.”

The post Why Deloitte Built a Tax AI That Knows When to Say ‘I Don’t Know’ appeared first on Analytics India Magazine.

The Department for Promotion of Industry and Internal Trade’s (DPIIT) working paper on the copyright use in generative AI is creating quite a stir.

The government body, proposing a mandatory blanket licence for AI training, has sought to strike a balance between granting AI companies access to content while also enforcing copyright laws, ensuring content creators receive statutory remuneration through a single-licence, single-payment mechanism.

While the Ministry of Electronics and Information Technology has endorsed DPIIT’s proposed framework, the strongest early pushback is coming from India’s technology industry, which argues that the framework misunderstands the mechanics of AI training and risks burdening an emerging ecosystem still taking shape.

Tech industry questions the core premise

Nasscom, India’s apex technology industry body, has reiterated its preference for a Text and Data Mining (TDM) exception with opt-out rights, which would enable copyright holders to oppose the use of their works in training models.

Replying to queries from AIM, Ankit Bose, head of AI, Nasscom, argued AI training involves “large-scale, automated, non-expressive processing of lawfully accessed data at the input stage,” which is fundamentally different from commercial exploitation.

As training does not produce expressive copies or substitute creative works, Bose believes treating training as a licensable act creates conceptual and practical problems.

For this reason, Bose sees opt-outs as a targeted, proportionate remedy. They give creators who are commercially sensitive or strategically exposed the ability to reserve their works from being used to train models.

He further argued that mandatory licensing at the training stage shifts copyright from its traditional domain of controlling use to controlling learning, a shift he believes goes beyond the intended function of copyright law and risks reshaping the incentives underpinning AI development.
Nasscom and Business Software Alliance, whose members include Google, Microsoft, Amazon Web Services, IBM, Salesforce and OpenAI, are among the entities that dissented to certain aspects of the proposal.

Innovation concerns

Bose also warned that the proposed blanket licence for any lawfully-accessed copyright-protected content could be “potentially burdensome,” particularly for startups, MSMEs and research-driven entities.

The industry body is concerned that the proposed royalty regime may introduce long-term cost unpredictability, since rates would be determined by a government-appointed committee with no clear precedent.

He also believes uniform royalty obligations applied across model sizes, dataset types and commercial scales would disproportionately affect early-stage innovators who rely on low-cost experimentation.

Treating training access as a paid statutory entitlement, Bose said, risks embedding fixed costs into the early phases of model development, where financial flexibility is critical.

Implementation challenges

Nasscom calls for a substantial redesign of the framework for it to be implemented.

Bose noted that attributing training influence to individual rights holders across India’s vast and often informal creative economy may be a monumental task.

He also questioned whether the proposed Copyright Royalties Collective for AI Training (CRCAT), a central body to set up royalties, could realistically achieve the transparency, auditing capacity and governance maturity needed to manage a national-scale royalty system.

He further noted that India lacks having a clear definition of AI training. The country also lacks a clear strategy on foundational model development versus fine-tuning, and how open-source or research models should be treated.

As India’s model diverges from the TDM-exception approach favoured in several global jurisdictions, including Japan, the UK, and the European Union, developers working across borders may face fragmented compliance requirements.

Bose said the system may only be implementable if it incorporates transparency-lite mechanisms, phased royalty applicability, startup exemptions, and explicit safe harbours for non-expressive training. Otherwise, it risks an “execution drag” that could eventually push innovation activity offshore.

Legal experts sound warnings

While the tech sector focuses on feasibility, legal experts say the proposal is on shaky statutory ground even before implementation challenges arise.

“India’s existing Copyright Act, 1957, does not permit a mandatory blanket licence for AI training,” said Priyanka S Kulkarni, senior legal advisor and solicitor specialising in IP. She explained that AI training “necessarily involves reproducing works in electronic form,” which falls squarely within the exclusive rights of copyright owners under Section 14.

None of the existing compulsory or statutory licensing provisions—Sections 31 to 32A and 31B to 31D—cover mass text-and-data mining or machine learning. These provisions, she emphasised, are narrow and tied to specific public-interest circumstances, and cannot be stretched to justify a sector-wide, automatic licensing regime.

Kulkarni added that the proposal preventing creators from withholding their works from AI training “would directly override their statutory exclusive right of reproduction,” making such a provision ultra vires (acting or done beyond one’s legal power or authority). It “would be vulnerable to legal and constitutional scrutiny unless Parliament first enacts a carefully structured amendment,” she cautioned.

Burden of proof and CRCAT

Some elements of the proposal, however, do find support in principle.

The proposed framework puts the burden of proof on the AI developer to establish compliance in case of any legal challenges. Anandaday Misshra, founder and managing partner at AMLEGALS, said it is “constitutionally sound under Article 21.”

He noted that the presumption aligns with Section 109 of the Bhartiya Sakshya Adhiniyam, which replaced the old Indian Evidence Act in 2023, places the burden on the party possessing exclusive knowledge. He called the presumption “a fair, proportional procedural mechanism.”

Misshra stressed, however, that CRCAT must rest on a strong statutory foundation. He said the entity would require its own chapter within the Copyright Act; clear, time-bound judicial review of royalty-setting to prevent arbitrariness; and distribution rules that ensure proportional and non-discriminatory treatment of all rights holders.

Drawing from India’s history with copyright societies, he said CRCAT would need mandatory third-party audits, a robust rate-setting framework, explicit penalties for non-compliance, and mechanisms to enforce obligations on foreign AI firms operating commercially in India.

Overregulation

Digital policy specialists also warn that the proposal assumes technical capabilities that today’s AI systems do not possess.

Jameela Sahiba, associate director at The Dialogue, a policy think-tank, said the recommendations “risk tipping the balance decisively towards overregulation in a sector that is still in its formative stages.”

She noted that the framework relies on advanced technical traceability that does not currently exist, and may introduce friction “precisely where agility and openness are most essential.” Monitoring training datasets, “massive, dynamic, and sourced from heterogeneous repositories” is operationally unrealistic, she added.

Sahiba argued that India must adopt a “technologically grounded, innovation-positive approach,” centred on dataset-level transparency rather than per-work accounting, which she said remains technically unfeasible.

Creators Deserve a Share

Filmmaker Joshua Sethuraman countered the tech policy narrative has long tilted toward large corporations, making the government’s attempt to rebalance power “crucial and much needed.”

He added there should be “no doubt whatsoever” that creators deserve royalties when their work is used to train or strengthen AI systems as they derive commercial value from human-created inputs.

“If a model’s capability is built on millions of creative works, the people behind those works deserve a share of the value,” he said.

Sethuraman also stressed that “the consumer of content reproduced by LLMs is often another creator,” making ethical use essential. Responsibility, he said, must rest both with users and AI developers, who must ensure their systems “respect the creative industry, not exploit it.”

As consultations begin, policymakers will need to determine whether India’s “one licence” vision can build cross-sector support, or whether the extensive legal, technical and institutional concerns raised by experts will force substantial redesign long before any amendment to the Copyright Act is drafted.

The post Can India’s AI Copyright Plan Survive Legal and Technical Scrutiny? appeared first on Analytics India Magazine.

When Lightbulb AI was founded in Mumbai in 2021 by Vishal Soni, Ritu Soni Srivastava, and Yogesh Sachdeva, the team wasn’t setting out to reinvent the global market research industry. In fact, their earliest experiments had little to do with advertising or consumer insights at all.

“Initially, Lightbulb came about as an idea in attention capture for the EdTech vertical,” Vishal Soni, chief product officer at Lightbulb AI, told AIM. The hypothesis was simple: as education moved online during the pandemic, could technology measure and improve student attention?

But as post-pandemic behaviour shifted and EdTech adoption slowed, the founders discovered an unexpected pull from the world of market research, where brands were struggling to understand not just what people said, but how they felt. Lightbulb’s emotion-tracking and eye-tracking models, built originally for children’s learning, suddenly unlocked groundbreaking new commercial applications.

“What happened after COVID was that the hypothesis on EdTech… didn’t pan out as we expected. But people wanted to understand how consumers were emoting when exposed to a particular stimulus,” Soni explained. This pivot proved decisive.

Today, Lightbulb AI stands at the forefront of insight technology, offering an integrated platform that combines eye-tracking, facial coding, AI-driven report generation, and generative AI-powered analysis.

And brands are taking notice, including Myntra, Merrill Research (US), and TVS Motors, with whom Lightbulb has collaborated. “We have submitted papers with these customer partners at the forums. So it’s on public record,” the founders confirmed.

Soni and Sachdeva’s partnership predates Lightbulb by nearly two decades. “We were together in our last venture also… and then we started this new venture,” said Sachdeva, who is the CTO.

This long-term collaboration helped them evolve rapidly into a deep tech research company. Their previous exit to RoundGlass also equipped them with operational maturity uncommon for early-stage founders.

Eye-Tracking, Facial Coding

Lightbulb’s platform blends quantitative and qualitative research capabilities in ways that traditional agencies rarely achieve. For quantitative work, the system analyses participants’ gaze patterns through eye-tracking, maps their emotional responses using facial coding, and incorporates survey data enhanced through generative AI. As Soni explained, “On the quantitative research aspect, we do eye-tracking… emotion mapping… and online surveys. All these things together give you the quant solution.”

On the qualitative side, innovation is even more fundamental. Traditional focus groups can take weeks to analyse, but Lightbulb compresses this entire process into under an hour. “A report writing which would take a week can now happen in 30 minutes… and presentations take another 10 minutes. Within 45 minutes to an hour, you’re done,” Soni noted.

Behind the scenes, AI continually improves accuracy. Sachdeva admitted that “Improving accuracy for emotions… other emotions have less raw data, and that is a major challenge,” but the team has trained its proprietary models on an enormous dataset of 9 million faces, making it one of India’s most robust emotion-recognition systems.

Given that Lightbulb processes video, facial expressions, and gaze data, privacy is paramount. Sachdeva emphasised that the company is fully compliant with international standards, including ISO 27001, ISO 27701, and GDPR. “No biometric information is stored, faces are mapped only to participant IDs, and data is deleted after six months to a year,” he explains.

Soni added that on the user-facing side, “We take explicit consent from the user… permission for camera, eye-tracking, and analysis at every stage.” This strict adherence to global standards has made Lightbulb a trusted partner for enterprise clients.

Global Recognition, and the Clients Behind It

Lightbulb AI has raised $1.5 million in seed funding from Chiratae Ventures. According to Soni, the team is not pursuing additional rounds immediately, choosing instead to strengthen product-market fit, stabilise recurring revenue, and move toward breakeven before exploring Series A opportunities. “We felt that we are capitalised enough to power through our Indian strategy,” he said, adding that the firm would look at fundraisers after another year, given its deliberate approach to sustainable growth.

Lightbulb AI’s rapid rise is reflected in industry accolades. The company became a finalist at MRSI and ESOMAR, won Gold at ESOMAR APAC for its work with Myntra, and was selected among the Top 50 deep tech startups at Singapore’s Slingshot Challenge. It was also one of only two Indian startups shortlisted at Hong Kong’s HKSTP startup competition.

These wins were driven by real deployments with clients such as Myntra, Merrill Research, and TVS Motors, spanning industries from retail to automotive to global research consulting. Each has used Lightbulb to decode human attention and emotion through advanced AI, demonstrating measurable improvements in insight depth and speed.

Market Momentum and the AI Productivity Mandate

According to Soni, “2025 and ’26 are watershed moments for productivity enhancement.” Companies across sectors now face pressure to do more with leaner teams, and Lightbulb’s value proposition—faster research, deeper insight, and lower operational cost—aligns directly with that mandate. He added that the industry-wide question today is, “Are you doing something with AI to enhance productivity?”

This demand is reflected in Lightbulb’s retention numbers. “We’ve had very little churn… some clients are in their third or fourth renewal cycle,” Soni noted, emphasising the stickiness of the platform.

Lightbulb began with a project-based model but has increasingly shifted toward a subscription paradigm. “The subscription model has worked really well for us… this year we’ve seen a lot of uptake,” said Soni.

Their growth strategy is unfolding in phases. After consolidating their presence in India, the team plans to expand into the US by deploying dedicated marketing and outreach personnel.

Simultaneously, they are building joint partnerships in Southeast Asia, tapping into established research networks. “We want somebody established there to take us as a technology provider along with them,” Soni explained.

The Challenges Ahead, and Why Lightbulb Isn’t Afraid

The founders are realistic about the competitive landscape. Sachdeva observed that as mainstream platforms like Google and OpenAI improve, and let people generate reports directly, they may not be interested in companies like theirs. Yet, he sees this as an opportunity, too: “Once they compare, they see the value we create beyond generic LLM outputs.”

Soni pointed to a larger existential challenge. “The speed at which OpenAI and Gemini are executing… that’s a major challenge for any organisation that relies on AI models,” he said, noting that large foundational models can generate code, images, videos, and analysis at a pace that reshapes entire industries.

Despite this, Lightbulb retains an advantage: deep specialisation in multimodal research workflows, something generic AI tools cannot replicate without domain-specific engineering, contextual tuning, and integrated gaze and emotion capabilities.

A Future Built on Multimodal Intelligence

Lightbulb’s next breakthrough aims to bring together three historically siloed data streams: eye-tracking, facial coding, and survey data. “We’re working on integrating these to bring out highly meaningful unified reports… that’s the challenge we’re really excited about,” Soni said.

This vision reflects a broader shift toward AI-native insight generation, where multimodal signals converge into an automatically generated narrative about consumer behaviour, richer, faster, and more accurate than anything created manually.

In four short years, Lightbulb AI has evolved from a pandemic-era EdTech experiment into one of India’s most advanced insight platforms, trusted by global brands, awarded by international research bodies, and recognised among deep-tech innovators.

Its founders are aware of the challenges ahead, but they remain grounded in the belief that understanding human attention and emotion at scale is a universal need, one that AI, when implemented responsibly and creatively, is uniquely positioned to solve.

If the last few years are any indication, Lightbulb AI isn’t just illuminating consumer behaviour. It’s lighting the path for the future of the research industry itself.

The post This Mumbai-based Startup is Illuminating the Future of Consumer Understanding appeared first on Analytics India Magazine.

IT services company Mphasis’ Sparkle Innovation Program has emerged as a significant channel for enterprise-focused innovation, helping it accelerate solution development, strengthen its AI-led portfolio, and support clients across banking, insurance, healthcare, logistics, and other sectors.

Launched in 2016, the initiative aims to foster innovation by partnering with startups, academic bodies and research entities. Over the years, Mphasis has partnered with ventures across sectors, including fintech Upswot to cater to US-based regional and global banks, as well as the Museum of Art & Photography through its CSR arm.

In an email interaction with AIM, Srikumar Ramanathan, chief solutions officer at Mphasis, said the programme is delivering measurable business outcomes and is set to expand further over the next two years.

Ramanathan said Sparkle has shortened evaluation cycles, reduced operational costs and increased pipeline velocity by enabling differentiated co-innovated offerings. He noted that several Sparkle-led engagements have matured into multi-year contracts across industries, including banking, insurance, healthcare, and supply chain.

The programme, he said, has helped clients adopt AI-driven, cloud-native and digital transformation initiatives more rapidly, creating new revenue streams, improving efficiency and enhancing competitiveness.

Through the program, Mphasis has been actively collaborating with Nasscom’s InnoTrek participants, particularly US-based startups, by offering mentorship, enterprise solution guidance, and customer access.
InnoTrek aims to help selected startups achieve global visibility and scale by supplying the necessary knowledge, tools, and connections.

In the 2025 edition of the programme, Mphasis recently partnered with five standout startups from the InnoTrek cohort—Edgeable AI, Perpetuuiti Technosoft, QuoQo, and SuperBryn AI—helping them refine their offerings and connect with enterprise clients worldwide.

The company said it will be working closely with these startups to co-create enterprise-grade solutions by providing technical mentorship, refining their go-to-market strategies, and facilitating access to global enterprise clients.

PoCs to Full-Scale Production

The CSO said a significant number of proofs-of-concept have moved into full-scale production deployments.

These span digital banking back-office automation, data intelligence, fraud detection, workflow digitisation, underwriting and claims-processing automation in insurance, patient data analytics and digital workflows in healthcare, as well as process automation and supply-chain visibility solutions in logistics and manufacturing.

He added that Sparkle also supports cybersecurity and IT resilience use cases, including threat detection, risk monitoring, and compliance automation.

Across sectors, Mphasis’ co-innovated solutions have delivered quantifiable results, Ramanathan asserted.

In insurance, cognitive and automation platforms improved underwriting and claims turnaround while lowering manual processing costs. In banking, AI-enabled automation reduced dependency on legacy systems and improved onboarding and service efficiency.

Healthcare clients, he said, saw better patient data management and more efficient care coordination. Logistics and supply-chain firms benefited from modernisation and productivity improvements, while cybersecurity clients improved their security posture and remediation cycles.

“Across these examples, clients saw faster deployment, improved efficiency, reduced manual effort, and better scalability,” he noted.

In Q2 FY26, the company reported new deal wins (TCV) worth $528 million, with 87% coming from new-gen services such as cloud, AI, and digital engineering. During the quarter, it scored six large deals, including one contract over $100 million with a BFSI client and two above $50 million, underscoring sustained enterprise-level demand.

This followed a strong Q1 FY26, where Mphasis logged $760 million in TCV wins, its highest quarterly figure in recent times.

Ramanathan said Sparkle acts as a feeder into the company’s AI-led, platform-driven portfolio.

By validating partner technologies in sandboxed environments and integrating them into Mphasis’ Front2Back architecture and X2C² framework for quick cloud adoption and cognitive transformation, the company has expanded its AI services more rapidly.

This includes document intelligence, voice AI, cognitive automation, and predictive analytics.

He said time from concept to deployment has “shrunk significantly,” enabling capabilities such as digital underwriting, cloud-native banking workflows, healthcare data platforms and supply-chain automation to reach production faster.

Selection Process

Mphasis selects partners based on alignment with client pain points, technical maturity, enterprise readiness, and integration capability. The CSO said scalability, cloud-native design, compliance standards and complementarity with Mphasis’ frameworks are essential criteria.

Once selected, technologies undergo technical due diligence, sandbox prototyping, use-case mapping and hardening before being integrated into Mphasis’ reference architectures and deployed with enterprise-grade support and SLAs.

Scaling innovations globally comes with challenges such as regulatory differences, legacy system complexity, performance scale-up, and security and compliance requirements. Mphasis addresses these through early compliance assessments, standardisation via its architectures, stress-testing, and leveraging the company’s global delivery footprint and domain-specific Centres of Excellence (CoE), Ramanathan said.

Sparkle differentiates itself by being “client problem–first” and embedding co-innovated solutions directly into enterprise delivery pipelines through Front2Back and X2C², he added.

He said the model is outcome-driven and enables future-ready, cloud-native, AI-led and compliance-focused modernisation for clients.

Ramanathan claimed the company is seeing strong enterprise traction in AI-led automation, cloud-native modernisation, cybersecurity and compliance automation, voice AI, and supply-chain and logistics automation. These areas reflect rising demand for resilience, scalability and better customer experience, particularly in regulated sectors.

Evaluating the programme, he said Sparkle has matured into a “dependable innovation channel” that bridges emerging technologies with real business needs.

Over the next 12–24 months, Mphasis expects to scale more PoCs to global production, deepen investments in AI-native and cloud-native platforms, expand domain-specific blueprints, strengthen global delivery CoEs, and support growing demand for integrated AI, automation, cloud and cybersecurity solutions.

“Sparkle is evolving into a robust, enterprise-ready, multi-industry innovation ecosystem that supports clients’ digital transformation journeys end-to-end,” he signed off.

The post How Sparkle Program Is Winning Multi-Year Deals for Mphasis appeared first on Analytics India Magazine.