Getting a single drug to market takes, on average, 10 to 15 years and somewhere between one and two billion dollars. That number isn't an outlier—it's the industry standard. And it still frequently ends in failure. The preclinical phase alone, before a molecule ever enters a human body, eats up four to six years and $100 to $200 million. Most of that time is spent running experiments, failing, going back to the drawing board, tweaking a molecule, and running them again.
This is the loop that pharma has been stuck in for decades. More data gets generated every year in biotech, but the process of turning that data into a viable drug hasn't fundamentally changed. Synthetic chemists still manually tinker with molecular structures. Labs still run hundreds of iterations to find the few that work. The cost keeps rising.
Ritvik Vipra wants to break the loop. His Bengaluru-based startup, Medvolt AI, is building a platform to compress the preclinical phase—not by replacing scientists, but by cutting the number of experiments they need to run to get to the right answer.
Fewer shots, better aim
Vipra is an IIT Roorkee technologist, not a biologist. His Co-founder and CEO, Dr. Madura, brings three decades of life sciences experience, a PhD in cancer biology and roughly 10 commercialized products across diagnostics and therapeutics. The pairing is deliberate. Medvolt's entire thesis depends on scientists and AI working in the same loop, not in parallel.
From the start, Medvolt built its platform with medicinal chemists and biologists in the room; not as reviewers at the end, but as collaborators throughout. The result is a system that can take a drug discovery project from hypothesis to shortlisted molecules, using generative chemistry and physics-based simulations to predict how candidate molecules will interact with target proteins in the body.
Vipra makes a simple pitch: instead of running 100 experimental iterations, Medvolt's platform helps a team get to the same outcome in 30. That's up to 3x reduction in project turnaround time and up to 10x reduction in cost.
Where OpenAI fits inside the stack
Medvolt's core generative chemistry models are trained in-house, on datasets the team has spent years curating and validating with domain experts. But OpenAI tools run across the platform in ways that are less visible and arguably just as important.
The knowledge discovery component, essentially the platform's literature and data intelligence layer, uses OpenAI embeddings as its foundation. These are then fine-tuned on Medvolt's own curated datasets to make them specific to biomedical use cases, and then power a retrieval-augmented generation pipeline that keeps the knowledge base continuously updated.
"We researched quite a bit on the agentic infrastructure side," Vipra says. "And we still arrived at OpenAI and their models, because those are the most trustworthy. Provided you give the correct context and your prompt engineering is sufficiently detailed, you get very good results."
GPT models remain their benchmark. The company uses OpenAI tools in daily activities across functions—developers debugging code, chemists improving workflows, and the marketing and business development team drafting content that has to hold up scientifically.
That last use case matters more than it might seem. Medvolt sells to scientists—not consumer audiences. The bar for credibility is high. "You're not marketing to a consumer audience. You need to have a baseline of marketing fundamentals, but if you're not scientifically robust, no one will take you seriously," Vipra says. ChatGPT, given the right context, bridges the gap.
Watch now
Planning in weeks, not months
One of Medvolt's recent collaborations is with ACTREC, the research division of Tata Memorial Centre, India's most prestigious cancer research institution. The work involves triple negative breast cancer, one of the harder subtypes to treat.
Medvolt has already shortlisted candidate molecules through its platform, and in vitro and in vivo testing is now underway at ACTREC. But getting to the starting line for those experiments, designing the experimental plan, creating documentation templates, aligning on timelines, is itself a significant piece of work.
"Traditionally, that kind of planning would require five or six domain experts working together for at least two months," Vipra says. "We did it in 15 days with one senior domain expert and two junior people, using ChatGPT to create detailed execution plans and templates."
The reason this matters is if your experimental design is flawed, you can't tell whether a bad result means your molecule was wrong or your test was. Bad planning at the start corrupts everything downstream. Getting it right quickly isn't just an efficiency gain; it's what makes the science trustworthy.
Delivering months of work in weeks
With a European client, a French subsidiary of a Chinese pharma company, Medvolt was handed an aptamer design project. The client needed a thorough literature analysis, key decisions made on the basis of it, and optimal molecular designs delivered. Their original expectation was four to five months.
Medvolt delivered in one month. This was a combination of Medvolt's knowledge discovery engine (powered partly by OpenAI embedding models), its generative chemistry platform, and faster client communication enabled by AI tools.
"If a complex client requirement comes in and it takes you two weeks just to understand and respond to it, you've already lost two weeks," Vipra says. "If you can go back to them in three or four days with the same resources, that compounds quickly."
The project also showed how speed translates to cost savings downstream. The client originally planned to synthesize and test 50 to 60 molecular designs. Medvolt gave them a refined shortlist of 20 to 25, confident enough in the filtering to halve the number of lab iterations required.
What it could mean for India and pharma
AI drug discovery is one of the hotter fields globally right now. The US and China are ahead; Europe is behind; India is catching up. The first fully AI-designed drug candidates are entering clinical trials. A few may reach patients within years.
For India specifically, Vipra sees a particular opportunity. Pharma R&D is capital intensive. India's ecosystem doesn't have the same depth of capital as the US or Europe. A technology that can genuinely reduce the cost of preclinical work—even by 25 to 50%, even well short of the theoretical 10x ceiling—changes what's possible for Indian research institutions and companies.
"Even a single percentage point improvement in accuracy or efficiency drives a whole lot of value in this sector," Vipra says. "If you reduce the cost of a drug from a billion dollars to 500 million, that's still saving 500 million. That's a huge amount."
Medvolt is working with the Indian government and is open to public-private partnerships to bring this capability into the domestic pharma ecosystem at scale. The loop that's defined drug development for three decades may finally be getting shorter.
Watch now
Original Article
(Disclaimer – This post is auto-fetched from publicly available RSS feeds. Original source: Yourstory. All rights belong to the respective publisher.)
