PopVax is an Indian full-stack biotech building first-in-class and best-in-class vaccines and cancer immunotherapies using machine learning-driven protein design and relentless empiricism. We design, develop, and manufacture our own RNA medicines end-to-end because we believe great pharmaceutical science can only flourish in tight feedback loops that iterate rapidly. PopVax’s experimental work and clinical dose production is based at the RNA Foundry, our integrated R&D and GMP-capable clinical dose production facility in Hyderabad.

PopVax's north star is our goal of developing novel vaccines and therapeutics over the next decade with the cumulative ability to save 1 million lives each year – the Million Lives Mission. To that end, we are developing first-in-class vaccines against Hepatitis C, Strep A, and adult pulmonary TB; broadly-protective best-in-class vaccines against COVID-19, influenza, malaria, and HPV; and precision immunotherapies against hard-to-treat solid tumours such as liver cancer and pancreatic cancer. Beyond existing diseases, we are leveraging our high-speed platform to build rapid-response biosecurity capabilities against engineered de novo pathogens.

Our mission is funded by Vitalik Buterin’s Balvi Fund, the Gates Foundation, the US Biomedical Advanced Research and Development Authority (BARDA), Renaissance Philanthropy, and Good Ventures, with individual investments from Enveda founder Viswa Colluru and Tesla self-driving AI pioneer Dhaval Shroff. Our first program, an open-source broadly-protective COVID-19 vaccine, will begin a Phase I clinical trial in Australia in mid-2026. This is just the start – we intend to advance 6+ novel vaccine and immunotherapy programs into human clinical trials over the next three years, decisively demonstrating that world-class biotech R&D is possible in India. 

No matter the job title, each person’s role at PopVax is ultimately about helping bring safe, effective new medicines that represent a step-change over the current standard-of-care to the people who need them, as quickly as possible. If you are looking for a place where the ambition is high, the learning curve is steep, and the work matters to billions, you’ll feel at home here.

If you’re excited by the idea of advancing scientific, clinical, and regulatory frontiers of vaccines and immunotherapies, spending each day developing medicines with the potential to save millions of lives, and building a generational global pharmaceutical company in India along the way – join us!

Role Overview

Here's a bet we're making at PopVax: the future of wet lab automation does not belong to the $500,000 liquid handler.

If you've ever worked in or around a biology lab, you've seen the machines. Highly specialised, exquisitely finicky, astonishingly expensive liquid handling robots that can pipette with superhuman precision — as long as you spend six weeks configuring them, hire a dedicated engineer to babysit them, and never ask them to do anything they weren't explicitly programmed to do. They are marvels of engineering. They are also brittle, inflexible, and wildly inaccessible to the vast majority of labs on the planet that can't afford half a million dollars for a single instrument.

We think there's a better way. We think that low-cost, general-purpose robotic arms — the kind that cost a few thousand dollars, not a few hundred thousand — combined with generative robot control models that can learn to execute complex scientific experiments from demonstration and expert guidance, will fundamentally change how wet lab science gets done. Not just at PopVax, but everywhere.

This is not a distant research vision. We are building this now. And we need someone to lead it.

That someone is our Robot Learning Engineer.

Why PopVax Is the Right Place to Build This

Most companies attempting to bring modern robot learning into the lab are doing it from the outside — AI teams building models in one building, biologists working in another, and a project manager in between trying to translate. That's not how this works at PopVax.

PopVax is a full-stack mRNA biotech company with over 100 people — most of them wet lab scientists — designing, manufacturing, and clinically testing novel vaccines and immunotherapies at a pace that makes the pharmaceutical industry uncomfortable. Our scientists run thousands of experiments. They generate enormous volumes of task-specific expert data every day — pipetting, plating, mixing, transferring, measuring — the precise kind of embodied, domain-specific demonstrations that robot learning models are hungry for.

And the people leading this effort are not newcomers to either robotics or lab automation.

PopVax's Founder & CEO, Soham Sankaran, was a computer science PhD student at Cornell working at the intersection of distributed systems and robotics before he dropped out to start Pashi, a Y Combinator-backed robotics and automation software company. He built robots as a kid in Mumbai, studied CS at Yale, and has spent the better part of a decade thinking about how software should control physical systems. The fact that he then pivoted to biotech and built a vaccine company from scratch is its own story — but the robotics instinct never left.

PopVax's VP of Programs, Darshit Mehta, spent time at Ginkgo Bioworks — one of the world's pioneering self-driving lab companies — working on automating high-throughput wet lab experimentation. He knows intimately what works and what doesn't when you try to automate biology, and he brings hard-won operational intuition about where the real bottlenecks are.

You will work closely with both of them. And you will sit right next to — not across a campus from, not on a video call with — experienced wet lab scientists who will be your collaborators, your domain experts, and your source of the richest training data you've ever had access to.

What You'll Actually Do

You will lead the development of PopVax's generative robot control models — systems that enable low-cost general-purpose robotic arms to autonomously execute complex scientific experiments in the wet lab.

This means building models that can watch a scientist perform a multi-step experimental protocol — the kind of thing that involves pipetting precise volumes, transferring liquids between plates, mixing reagents in specific sequences, operating instruments, and making real-time adjustments based on visual feedback — and learn to do it. Not through painstaking manual programming of every motion and every parameter, but through the kind of generative, learned control that modern transformer and diffusion-based architectures are beginning to make possible.

Specifically, you will:

• Build generative robot control models. Design, train, and deploy models that take visual observations and task descriptions and output robot actions — the core loop of autonomous lab execution. You'll draw on modern architectures from the vision-language-action (VLA) model literature, diffusion-based policy learning, and related approaches to build systems that can generalise across the diverse and messy reality of wet lab tasks.

• Develop robot world models for the lab. Build predictive models of the lab environment — what happens when a pipette tip enters a liquid, when a plate is moved, when a centrifuge lid is closed — that allow the robot to plan, anticipate, and recover from the unexpected. Wet labs are not warehouses. The physics is fiddly, the lighting changes, the liquids behave in ways that don't always cooperate with simulation. Your world models need to handle that.

• Design the data pipeline from bench to model. You'll be sitting next to scientists who perform expert-level experimental protocols every single day. That is an extraordinary source of training data — but only if you build the infrastructure to capture, annotate, structure, and learn from it. You'll design the data collection strategy, the recording setup, and the pipeline that turns raw demonstrations into usable training signal.

• Close the loop between model and reality. The model has to work on a real robot, in a real lab, manipulating real liquids and real labware. You'll own the full stack from model training through sim-to-real transfer (where applicable) to deployment on physical hardware. When the robot drops a plate, you'll figure out why and fix it — in the model, in the data, or in the control loop.

• Collaborate deeply with wet lab scientists. This is not a role where you disappear into a GPU cluster for six months and emerge with a paper. You'll work with scientists daily to understand what protocols need automating, what the failure modes are, what "good" looks like in their domain, and how to build systems that they actually trust to run their experiments. The scientists are your co-designers, not your end users.

Who You Are

You have experience building image or video generation models using modern transformer and diffusion-based architectures. You understand the generative modelling stack — the training procedures, the architectural choices, the data requirements, the failure modes — at a deep, hands-on level. You've trained large models yourself, not just fine-tuned someone else's.

Ideally, you've worked on vision-language-action models, robot foundation models, or robot world models — the emerging family of approaches that treat robot control as a conditional generation problem. If you've published in this area, great. If you've built systems that actually work on physical hardware, even better. If you've done both, we really want to talk to you.

You're comfortable at the intersection of machine learning and physical systems. You understand that the gap between a model that works in simulation and a model that works in a real lab is not a minor implementation detail — it's where most of the hard engineering lives. You're the kind of person who gets excited, not frustrated, when the real world refuses to cooperate with your model's assumptions.

You don't need a biology background. But you should be genuinely curious about what happens in a wet lab, willing to learn the difference between a 96-well plate and a 384-well plate, and excited by the prospect of building AI systems for a domain where the stakes — better vaccines, faster drug development, lives saved — are as high as they get.

Qualifications

• Bachelor’s, Master's, or PhD in Computer Science, Robotics, or a related field — or equivalent research and engineering experience.

• Strong hands-on experience building and training generative models (transformers, diffusion models, or related architectures) for image, video, or action generation.

• Solid software engineering skills in Python and deep learning frameworks (PyTorch or equivalent).

• Experience working with real robotic hardware or physical systems is strongly preferred.

• Published research or demonstrated project work in VLAs, robot learning, world models, or embodied AI is a significant plus.

Preferred (but not required)

• Experience with robot manipulation, particularly in contact-rich or deformable-object settings (liquids, soft materials, precision placement).

• Familiarity with sim-to-real transfer, domain randomisation, or related techniques.

• Experience building data collection pipelines for robot learning from demonstration.

• Exposure to laboratory automation, biotech, or scientific instrumentation.

Reporting Structure

You'll report to Darshit Mehta, VP of Programs, who worked on automating high-throughput wet lab experimentation at Ginkgo Bioworks and knows firsthand what it takes to make robots useful in biology. You'll also work closely with Soham Sankaran, Founder & CEO — a lapsed roboticist who dropped out of Cornell's CS PhD program to start a Y Combinator-backed robotics company before pivoting to biotech, and who has been waiting for this particular convergence of AI and lab automation for a very long time. And you'll collaborate daily with PopVax's wet lab scientists, who will be your partners in defining the problems, generating the training data, and validating the results.

This is a rare opportunity to work at the intersection of frontier AI and frontier biology, in a company where both are happening under the same roof. If you want to build the generative models that teach robots to do science — and you want to do it somewhere that the science actually matters — we'd like to hear from you.