Год: 2025

OpenAI has entered into a definitive agreement to acquire neptune.ai, a move positioned to strengthen the company’s internal tooling for frontier-model research. The acquisition focuses on improving how researchers observe, analyse, and iterate on large-scale model training.

According to OpenAI, developing advanced AI systems depends heavily on understanding how a model evolves during training. Neptune’s platform provides experiment tracking, run comparison, and real-time monitoring, giving teams clearer insights into complex model behaviour as it unfolds.

Neptune has collaborated closely with OpenAI in recent years to build tools that let researchers compare thousands of training runs and inspect metrics across layers. OpenAI said the team’s depth in this niche will help accelerate experimentation and improve decision-making throughout the training pipeline.

“Neptune has built a fast, precise system that allows researchers to analyse complex training workflows,” said Jakub Pachocki, chief scientist at OpenAI. He added that the company plans to integrate Neptune’s tooling deeply into its training stack to enhance visibility into how models learn.

Piotr Niedźwiedź, Neptune’s founder and CEO, called the acquisition “an exciting step”, noting the company’s longstanding belief that strong tools enable better research. Joining OpenAI, he said, brings that mission to a much larger scale.

OpenAI stated that it is looking forward to building “the next chapter of training tools” together with the Neptune team.

The company has recently declared internal ‘Code Red’ as competition from Google, DeepSeek and Amazon intensifies, prompting the company to prioritise new reasoning models over other projects. OpenAI is reportedly developing a model called Garlic, expected to rival Gemini 3 and Anthropic’s Opus series, with early results suggesting a potential GPT-5.2 or GPT-5.5 release in 2026.

Despite technical setbacks and questions over its scaling strategy, OpenAI maintains confidence in large-scale pre-training and is rebuilding capabilities in core model training. With strong user adoption, major compute partnerships and projected $20 billion revenue, the company is betting that renewed focus on scaling and reasoning will keep it ahead in the AI race.

The post OpenAI To Acquire Neptune, a Startup That Helps Train AI Models appeared first on Analytics India Magazine.

Over the last two decades, India’s startup narrative leaned on a familiar axis: engineering institutes produced technologists, technologists started companies, and the labs and dorm rooms in those institutes fed the country’s innovation pipeline.

Management schools, however, were often the backbenchers in this conversation. But, that scenario is evolving now. The Indian Institute of Management (IIM) Bangalore, IIM Lucknow, and others have built incubation centres that not only accelerate market-ready startups, but also train researchers to become founders.

They equip companies with model training and data pipelines, while also training on pricing, compliance, sales cycles, manufacturing challenges, stakeholder psychology, fundraising, and the slow, painful art of scale.

Having become nurseries for deep tech and AI ventures, these incubators are complementing IITs by introducing managerial discipline to back engineering vision.

Business Incubation Finds a New Frontier

Anand Sri Ganesh, CEO of Nadathur S Raghavan Centre for Entrepreneurial Learning (NSRCEL) at IIM Bangalore, said that for any tech or IP-driven innovation, the intervention of a business incubation is critical mainly in two areas.

The first occurs “somewhere in the TRL three, four stages,” when “innovation is starting to crystallise within the lab.”

This is where IIM incubators step in, long before a product hits the market. They ask founders to consider interoperability, price thresholds, customer archetypes, certification, procurement, and downstream integration at a time when most scientists are still polishing their core technology.

“The credibility of being part of a premier incubation programme builds trust and encourages enterprises to engage with us under NDA,” said Raja Mohan, founder and CEO of Prodloop, a voice AI startup incubated at NSRCEL.

NSRCEL provided peer networks, industry connects, and compliance playbooks that accelerated enterprise pilot conversations. For a young AI company, that support is catalytic.

It also helped Prodloop navigate integration with legacy call-centre stacks, a task more complex than many engineers realise. Enterprises often run “heterogeneous or legacy telephony stacks that lack standardised APIs,” Mohan noted.

Compliance thresholds such as ISO 27001 and SOC 2 further slow down onboarding. NSRCEL’s grant support helped Prodloop pursue certification faster and exposed the team to seasoned SaaS founders with real-world integration experience.

However, the second moment where business incubation is critical comes after the prototype is real enough to be demonstrated. “Once beyond the lab, the invention gets into a customer demonstrable prototype pre-MVP ready,” Ganesh said.

“That is when the real translation of the business model takes shape,” he added. This is where most AI ventures stall, not due to technical incompetence, but because few engineers have ever sold to an enterprise buyer, or structured a stakeholder-driven pilot.

Prodloop experienced this friction firsthand. Mohan isolated the core barrier to training robust voice models: “Access to diverse, high-quality conversational datasets… is critical.”

But large organisations are “still hesitant to share voice data for POCs due to compliance concerns, even when PII/OII masking and anonymisation is offered,” he added. Public datasets are inadequate, scarce, fragmented, and unrepresentative of enterprise reality.

These blind spots are precisely what IIM incubators target.

According to Ganesh, one of NSRCEL’s core evaluation criteria is whether founders have “sweated the idea.” They must show they have read papers, met suppliers, spoken to customers, and challenged their own intuition.

IIM Lucknow’s Role

The Enterprise Incubation Centre (EIC) at IIM Lucknow is another nursery for innovative aspirations.

When asked what most AI companies struggle with, Amrit Tiwari, head of investment at EIC, didn’t reach for buzzwords. “AI startups often struggle with data privacy compliance, model explainability, and domain-specific validation.” However, these are not model-architecture problems; they are industry problems, he said.

EIC intervenes through expert-led regulatory workshops, connects startups with legal advisors, and facilitates pilot testing, Tiwari said. It guides them through compliant data pipelines, ethical AI frameworks, and enterprise integration, solving the bottlenecks that engineering institutions rarely deal with.

Durability carries more weight than hype cycles at EIC. “We evaluate the startup’s tech adaptability, modularity, proprietary IP, data advantage, and model scalability,” Tiwari said. The goal is defence against obsolescence in a rapidly shifting AI landscape. Instead of relying solely on external LLMs and APIs, IIML EIC favours “problem-first, model-agnostic solutions.”

CallerDesk is a case study in how an incubator can change a company by reshaping its thinking. When co-founder Kaushal Bansal began the startup in 2016, he saw a gap in India’s voice communication market. There was no tool SMEs could use with the frictionless simplicity of WhatsApp or Dropbox.

For telephony, founders had to choose between a ₹10,000 IVR system or a ₹50 lakh call centre infrastructure. “We thought that we should start this thing,” Bansal recalled. “They [enterprises] can just download the app, configure the IVR, and onboard their cloud call centre things at 90% discounted rates.”

Tiwari sees a repeating pattern in most AI startups’ failure: they stall after pilots. Enterprise clients hesitate due to integration challenges. Teams lack a go-to-market strategy. Data quality collapses. EIC bridges this gap by providing corporate partnerships for paid proofs of concept, investor access, and deployment advisory.

CallerDesk embodied a scrappy ingenuity common to Indian entrepreneurs. But as Bansal admitted, it lacked direction. “Before IIM Lucknow, we were…very misaligned.”

At the EIC, the founders realised what they had built was “actually very good,” and that it was “time to scale.” EIC provided CallerDesk mentorship, infrastructure, workspace, and funding, including ₹25 lakh in seed support.

The change was not theoretical. CallerDesk’s client base grew threefold, and call volume increased fivefold. The company went from dealing with a handful of enterprise clients to supporting “around 50,000 plus agents… on a daily basis.” Its revenue mix matured into roughly 30% subscription fees and 70% usage-based charges.

The Convergence of IIMs with the IITs

When Ganesh was asked whether IITs or IIMs have the advantage in nurturing startups, he rejected the idea of rivalry entirely. “The multiplier effect of joint incubation is very high,” he said.

That collaboration brings deep IP development and testing capabilities, along with ability to transfer to markets and create entrepreneurial mindsets, he emphasised.

In other words, India’s next generation of innovation belongs to avenues where labs and markets converge.

Ganesh denied that IITs, being technical institutes, hold an inherent advantage. For him, co-incubation is the most potent form of support, as engineering campuses alone can not handle the whole innovation lifecycle.

To achieve this, NSRCEL works with a national network of technical institutes. They work with IIT Guwahati on circularity and climate tech, IIT Hyderabad on medical equipment, IIT Madras on climate and deep tech, and IIT Kanpur on robotics and avionics. These partners provide “anything from design, fabrication, prototyping, testing, validation, and pre-certification,” said Ganesh.

Meanwhile, NSRCEL complements this with business incubation, focusing on product-market fit, venture readiness, stakeholder management, and sales discipline.

However, the work gets messy. “Giants find it difficult” to collaborate, Ganesh admitted, referring to larger institutions and corporations. True co-incubation “is more embedded in design, difficult to execute, but more effective if you’re able to execute it.”

Management Meets Engineering

India is now at an inflexion point. Engineering institutions continue to play a vital role in creating the intellectual property that drives deep tech innovations. At the same time, management incubators are essential in developing the commercial, regulatory, strategic, and operational frameworks necessary for the sustainable success of these technologies.

Summing up the landscape, Ganesh said that in India, “the incentives are not yet aligned” between researchers, corporates, startups, and capital. “I’m hoping it will happen in time.”

But, the change is indisputable. CallerDesk wants to be “the WhatsApp for voice in India.” Prodloop is expanding multilingual dataset coverage at enterprise scale. IIM Lucknow is training AI companies to withstand regulatory scrutiny and commercial pressure. NSRCEL is turning students and graduates into entrepreneurs.

The old innovation axis has evolved, as IIMs, instead of competing with IITs, complement their innovations.

The post IIMs Are Quietly Incubating AI Startups That Will Outlast the Hype appeared first on Analytics India Magazine.

Consulting firms rarely make headlines about AI. But when they do, it tends to be either about multibillion-dollar investments or a high-profile mishap linked to their use of AI. But looking at it as a whole, it is clear that the consulting industry’s approach to AI has plenty of cracks.

It is not a collapse, nor a downfall. It feels more like a pressure test. For three decades, the Big Four sat at the centre of every major transformation. Deloitte, PwC, EY and KPMG, along with Accenture, McKinsey or BCG, built vast engines that combined audit weight, global processes and armies of analysts. For years, that model felt unshakable.

Nishant Pahwa, connection and network manager at Cresent Core Consulting, with prior experience at EY and PwC, offered a telling example. He recalled that in early 2025, two CXOs walked out of a boardroom when someone remarked, “Deloitte gave a great deck… but I need someone who can solve this by next Friday, not next quarter.”

That sentiment is now widespread. Many companies would rather turn to smaller firms that adopt and deploy AI far more quickly, rather than spending thousands of dollars on big firms and waiting for quarterly results. IBM’s consulting head, Mohamad Ali, recently told The Times of India in an interview that consulting firms’ model is under threat.

“The future of consulting is going to be a hybrid of people plus software,” he said, adding that the companies that do not adopt this would fail.

The firms are not shrinking. Their numbers still climb. Deloitte stands at more than $70 billion in FY25, PwC at $57 billion, EY at $53 billion and KPMG at around $38 billion. But the market around them is changing faster than they can turn.

But, What is Actually Happening?

Clients want less analysis and more execution, less manpower and more speed. AI has pushed this shift forward at a pace that has surprised even insiders. Business Insider recently highlighted how smaller firms, which are basically consulting startups using AI, have become the biggest competition to the Big Four consulting firms.

Every large firm has pushed aggressively on AI. Deloitte built the Deloitte AI Institute, PwC announced a billion-dollar commitment, EY launched EY.ai, and KPMG tied up with Microsoft and AWS for large AI programs.

Accenture is a particularly interesting example. After spending more than $3 billion on AI, the firm described the returns as underwhelming. The company is now also rebranding its 8,00,000 workforce as ‘reinventors’ to adapt with AI and has announced a partnership with OpenAI to integrate ChatGPT Enterprise to all its employees.

In many ways, this mirrors the forward-deployed engineer model that Palantir has followed for a long time. These are essentially technology consultants who will advise companies while sitting within them, hinting strongly about where the industry is heading.

While the large firms continue to handle the multi-year projects, a different race is unfolding in the shorter cycles.

Boutique firms like Xavier AI, NextStrat, Consulting IQ, Perceptis, and a few others have stepped into high-speed, high-depth mandates. These firms win because a team of 12 specialists outpaces a team of 200 generalists inside a Big Four engine.

Ironically, these firms are built by several former McKinsey consultants. Meanwhile, McKinsey itself recently cut around 200 jobs to focus more on AI-related roles. This is also what went down with Accenture, which announced a slew of layoffs in its latest quarterly report.

Yet, the challenge is that the struggle continues. In Canada, The Independent found that a Deloitte report for Newfoundland and Labrador contained citations that did not exist or did not support the claims made. The report cost around $1.6 million Canadian.

This followed its first failure in Australia, where Deloitte refunded part of a A$459,000 contract because the report had fabricated references created by an Azure OpenAI system using GPT-4o. Australian senator Deborah O’Neill said, “Deloitte has a human intelligence problem.” She said government agencies might be “better off signing up for a ChatGPT subscription”.

Deloitte said the errors did not change the findings. But the episode shook trust across the sector.

The AI Hard Sell

This is what makes the current friction so interesting. The consulting world is selling AI harder than ever, and failing to do so.

KPMG went from zero to $650 million in revenue from generative AI in one year. Deloitte launched its $3 billion Zora AI with NVIDIA, focused on autonomous AI systems. BCG now earns about 20% of its revenue from AI projects. McKinsey offers more than 140 AI accelerators. The story they present to clients is clear. But clients are starting to notice something different.

Many of the tools these firms sell are now directly accessible to companies through OpenAI, Anthropic or even Google and Microsoft. They sit on cloud platforms and behind APIs. They come from the same model providers everyone uses, alongside forward-deployed engineers.

There is also a deeper problem. Enterprises are not ready for AI at the scale consultants promise. Mukesh Bansal, founder of Nurix AI, earlier said, “Everyone is building AI agents, and yet so few AI agents are in production doing real work.”

He said agentic AI companies need to operate like a mix of McKinsey and Infosys, pairing strategy with execution. Boutiques have stepped into this gap with AI native models. As Business Insider first reported, many of these firms are built by former MBB or Big Four consultants who wanted less bureaucracy and more speed.

They serve clients who could never have afforded a McKinsey team. Their growth is rapid. FT reported that Xavier AI says revenue is doubling month over month. Perceptis raised $3.6 million, SIB has identified more than $8 billion in savings, and Genpact cut $40 million in costs with AI through its Client Zero programme.

These firms show a version of consulting that feels more like a product—faster and easier to consume.

The consulting world is now a mix of giants, boutiques and AI native players. The cracks are not signs of collapse. They are signs of recalibration. The winners will be the ones who can combine real human judgment with AI in a way clients can trust. The old fortress is still massive, but the drawbridge is no longer one-way.

The post AI Was Meant to Save Consulting. It Exposed the Cracks Instead. appeared first on Analytics India Magazine.

Yoodli, a US-based AI-powered learning platform, has raised $40 million in its Series B funding round led by WestBridge Capital, with participation from Neotribe and Madrona.

Founded in 2021 by Varun Puri and Esha Joshi, Yoodli has secured Series B funding just a few months after announcing its Series A round in May. To date, the company has raised nearly $60 million.

The startup’s platform uses AI to simulate real-world scenarios, from sales calls and leadership coaching to interviews and feedback sessions, giving users instant, personalised feedback they can practice privately and repeatedly.

This funding will accelerate Yoodli’s investment in AI coaching, analytics, and personalisation while expanding its reach across enterprise learning, GTM enablement, and professional development. The company plans to grow its product, AI research, and customer success teams as it continues to scale globally.

“This round helps us scale our team and serve more enterprises on a true end-to-end experiential learning platform,” said Varun Puri, cofounder and CEO of Yoodli. “We’re reducing the time it takes to acquire real skills, ensuring employees are ready for game time, and saving organisations countless hours lost to passive coaching.

The Yoodli platform is used by major companies like Google, Snowflake, Databricks, RingCentral, and Sandler Sales. Adoption has surged due to a workplace shift towards experiential learning, allowing employees to improve through guided practice rather than just consuming training content.

“We see Yoodli defining a new category of AI-native learning tools for the enterprise, as companies today seek scalable, AI-driven solutions to train and upskill their workforces. The Yoodli team has built a platform that brings a high level of precision and scalability to skill development, and we’re excited to partner with them as they scale,” said Manthan Shah, principal at WestBridge Capital.

The post AI Learning Startup Yoodli Raises $40 Million in Series B Led by WestBridge Capital appeared first on Analytics India Magazine.

KPIT Technologies has pursued a series of tightly scoped acquisitions and minority investments over the past few years to assemble a comprehensive software-defined vehicle (SDV) and mobility ecosystem.

While the moves appear varied at first glance, ranging from in-vehicle networking firms to digital experience platforms, the through line is deliberate.

The company has been aiming to deepen essential technology capabilities and expand the value it can deliver to global OEMs and partners navigating the transition to software-led mobility.

Elaborating on the approach, Mohit Kochar, chief marketing officer at KPIT, told AIM that the acquisitions “collectively strengthen its software-defined vehicle capabilities, deepen domain expertise, and enhance its long-term AI-enabled engineering roadmap.”

He pointed to investments in Caresoft Global Engineering’s Solutions business, N-Dream (AirConsole), FMS, PathPartner, Technica and a set of smaller minority stakes.
KPIT’s early additions like PathPartner and FMS strengthened embedded engineering and digital experience foundations.
The Technica acquisition expanded in-vehicle networking and test infrastructure depth; the stake in N-Dream (AirConsole) added next-generation in-car digital experience capabilities.
The purchase of Caresoft Global Engineering’s Solutions business broadened engineering scale and customer access; and the recent investment in helm.ai advanced the company’s AI-enabled mobility roadmap.

According to Kochar, these decisions follow a consistent two-fold intent: reinforcing capabilities critical to SDV programs and widening the scope of value KPIT brings to global OEMs and ecosystem partners.

In certain cases, Kochar said, the investments also create stronger strategic access to customers and open new areas of engagement.

Rather than approaching SDV as a monolithic shift, KPIT’s strategy distributes capability-building across the stack, aligning each acquisition with a specific engineering need that OEMs face, as vehicles become increasingly defined by software, data and AI-enabled functions.

Kochar underlined that autonomous driving is only one application within the wider SDV landscape.

He noted that KPIT has developed mature, organic capabilities in this space over many years as a strategic engineering partner to global OEMs.

Among the recent additions, Technica stands out for its deep expertise in automotive Ethernet, hardware, test infrastructure and reference architectures. These capabilities are essential for modern SDV programs in which reliable, high-speed data flow across ECU and zonal architectures becomes non-negotiable.

Other acquisitions similarly broaden KPIT’s ability to deliver differentiated software, systems engineering and next-generation digital experiences across mobility platforms.

Internally, each acquisition within the KPIT Group operates in a way that maximises customer outcomes rather than being forced into a uniform integration model.

Some maintain a high degree of independence, while others align closely with KPIT’s practices, delivery mechanisms and engineering processes.

Kochar emphasised that the common thread across them is customer value and execution excellence, not structural conformity.

AI Capabilities

This approach also extends to AI. KPIT positions AI and generative AI as central to its long-term roadmap, embedding them across the automotive software lifecycle to improve productivity, speed, quality and efficiency.

Acquired entities both benefit from and contribute to this ecosystem, but Kochar made clear that none function as isolated AI centres; instead, they operate within a shared technology framework.

The company’s differentiation, as it presents it, is grounded in deep domain expertise in mobility, long-standing OEM partnerships and its ability to translate advanced technologies, including AI, into real-world engineering impact.

KPIT’s strategy is not positioned around specific tools or platforms but around its capacity to scale complex automotive programs in markets where the shift to SDV architectures demands consistency, reliability and safety across multiple software layers.

In its Q2 FY26 results, KPIT reported revenues of $181 million, reflecting 4.4% year-on-year growth in dollar terms and 7.9% in rupee terms. The company also reported EBITDA margin expansion to 21.1% and a total contract value of new engagements worth $232 million.

Commenting on the performance, Kishor Patil, co-founder, CEO and MD, KPIT, described the quarter as one that strengthened the company’s foundation for the SDV transition.

He cited strategic investments including the closure of the Caresoft Engineering Solutions Business acquisition in Q2, the increase in stake in N-Dream and the investment in helm.ai in Q3 as building blocks aligned with industry direction.
However, the company did not address specific questions related to internal AI integration, data unification across acquisitions, AI safety workflows, generative AI usage in engineering, or the existence of a unified MLOps framework.

The post KPIT’s Big Bet: Buying the Brains of the Software-Defined Car appeared first on Analytics India Magazine.

Bangalore-based digital wellness platform Healthify has launched Ria Voice, a real-time AI health coach that uses OpenAI’s Realtime API to deliver natural, speech-to-speech coaching.

The global rollout makes Healthify one of the first companies to bring a fully multimodal, audio-native health agent into production, enabling users to speak to the coach, show it their meals or activity, and receive instant, context-aware guidance.

Ria Voice represents the 3.0 evolution of Healthify’s AI stack and moves beyond older voice assistants that depended on transcribing speech before generating a response. Operating natively in audio allows the system to detect emotions, handle diverse accents, support code-mixed languages, and respond with minimal latency.

The coach draws on nutrition, fitness, sleep, stress, glucose, heart rate and body composition data, enabling a level of personalisation that mimics human coaching. Users can log food by describing their meal or pointing their camera at a plate, eliminating the friction of manual entries.

“Our mission at Healthify has always been to put a personal coach in your pocket. With Ria Voice, we’re finally able to bring that vision to life at a global scale,” Tushar Vashisht, co-founder and CEO of Healthify, said. “By leveraging OpenAI’s Realtime API, Ria can understand your food, fitness, sleep, stress and metabolic data in one place and respond with human-like speed.”

Pragya Misra, head of strategy and global affairs, India at OpenAI, said the product reflects what OpenAI’s new real-time infrastructure was designed for. “Our Realtime API can transform voice AI from a static interface into a conversational experience. Healthify has used this capability to build a coach that is fast, responsive and capable of the nuance required for personal wellness,” she said.

Ria Voice supports more than 60 global languages, including over 14 Indian languages, and is trained on hundreds of millions of real-world conversations between Healthify customers and human coaches. It can also generate customised diet plans based on preferences, allergies and macro goals, expanding its role beyond logging and recommendations.

The company is making the experience available not just on its app but also through WhatsApp and wearable devices. Through integration with Ray-Ban Meta smart glasses, users can talk to Ria hands-free and track meals via photos captured directly on the device—making Healthify one of the earliest health apps built for Meta’s smart glasses ecosystem.

The post Healthify & OpenAI Launch Ria Voice, Realtime Multimodal AI Health Coach appeared first on Analytics India Magazine.