Technical consultant for early-stage deep-tech startups
When money, people and time are all short, I help founders build the product, the team and the strategy anyway. Unconventional routes where they pay off, no shortcuts on customers or compliance.
Four ways to work together. Each one applies the same approach: get further with the people, money and time you actually have.
Fractional CTO for the stage before you can hire one. Product and architecture decisions, hiring and building the engineering team, the technical story for your next round.
Read moreFor investors and founders before a round or acquisition. Code, architecture, team and claims, checked by someone who has shipped the kind of product being pitched. Three depths.
Read moreBefore product-market fit: what to build first, what to buy, what to skip, and who to hire. Short sprints with a decision at the end.
Read moreFeasibility and system design for agentic AI and robotics products, with safety and regulatory readiness built in from the start rather than bolted on.
Read moreThe plan is built for the runway you have, not the one you wish you had.
When the straight road is closed, there's usually another one. Finding it is the job.
Shortcuts that reach the customer aren't shortcuts. Everything else is fair game.
CE, ISO and EU AI Act readiness designed in early, so it costs weeks, not a rebuild.
I'm Abhishek Nair, a mechanical engineer turned robotics and AI builder, based in Berlin. I've co-founded two startups, led engineering and compliance in regulated robotics, and shipped AI systems in production. I run Padawanabhi, a small consultancy for founders who have to get a lot done with a little.
More about me
AI/ML
Four frontier AI models shipped in seventeen days this July, and my stack needed one config change. What absorbed the churn, why felt productivity lies, and when a model release still deserves attention.
AI/ML
After metrics kept missing degenerate gaits and LLM-iterated rewards hit the survival cliff, we tried a vision-language model as the fitness scorer. The VLM was harsher than metrics and produced actionable failure descriptions — and it scored a collapsing robot 62/100. The case study in honest fitness, plus the four-layer evaluation stack we landed on.
AI/ML
We ran NVIDIA's Eureka methodology — an LLM iteratively proposing reward functions for an RL agent — on cold-start humanoid bipedal walking. 24 candidates across two rounds, all fell by step 70. The LLM was thoughtful; the survival cliff was state-space coverage, not reward design.
Tell me what you're building and what's in the way. I reply within one business day, usually with a first thought, not a sales pitch.