AI4Science.Studio · by Hugo Penedones
AI where the physics matters.
Partnering with biotech, pharma, materials and engineering teams to solve hard machine learning problems.
How it works ↓
Hi, I'm Hugo Penedones. I have 20 years of ML engineering at the frontier, in science and in production, at scale. I was an early member of the AlphaFold team at Google DeepMind, the system behind the 2024 Nobel Prize in Chemistry. Later, I co-founded and led engineering at Inductiva, a cloud-HPC platform for simulation engineers.
Where I can help.
Biotech, Pharma & Materials
Machine learning for molecules and materials, from biological sequences to property prediction and generative design.
DNA/RNA Sequence Models · Structure Prediction · Materials Property Prediction · Generative Models · Machine-Learned Force Fields / Interatomic Potentials
Engineering, Manufacturing & Robotics
Fast surrogate models, physics-informed machine learning, and reinforcement learning for control, optimization, and design under real-world constraints.
Physics-Informed ML · Neural Operators · Design Optimization · Control & RL · Sim-to-Real · Surrogate Modelling
Currently accepting inquiries for upcoming project partnerships.
A focused studio.
One project at a time.
When I take yours on, it gets my undivided focus, not a fractional slice split across a dozen accounts.
Direct ownership.
If I need to scale up, I recruit a small group of scientists and ML engineers, but the intellectual lead is always mine.
Small Wins, Compounding.
Engagements run in 2-3 week research cycles. Each one delivers a concrete milestone, steadily reducing uncertainty.
Premium model.
Elite ML work is not cheap. You are paying for judgement that compresses timelines, not just execution.
AI-native engineering.
Coding is no longer the bottleneck; deeply understanding the problem and its constraints is. My value lies in pruning the search space and providing the intuition to skip dead-ends while leveraging AI to execute at 10x speed.
Human at every step.
I use the most advanced tools available, but you work directly with me. Just honest conversations, genuine curiosity about your problem, and work that's actually enjoyable to do together.
Typical engagements.
Long enough to do serious work. Short enough to stay focused. Projects don't drag on indefinitely.
Every cycle has its own budget and a clear go/no-go decision. You invest incrementally as confidence grows, not all upfront.
Scoping
Define the problem precisely, agree on success metrics, identify risks, and establish what a good outcome looks like. This phase ends with a written specification. No ambiguity.
Deliverable: technical spec & risk assessmentBuild
A series of 2-3 week research cycles. Early cycles are exploratory and answer feasibility questions; later cycles harden the system toward production. Each cycle delivers a tangible result and ends with a go/no-go decision before the next one starts.
Deliverable per cycle: working code, results report & next-cycle recommendationPublication
OptionalWhen the work sits close to the research frontier and you want to make it public, I can lead writing it up for peer-reviewed publication.
Deliverable: submitted manuscriptLet's talk about your project.
I'm selective about the projects I take on. I want to work on things that are genuinely hard and where ML can make a real difference. If that sounds like your situation, I'd love to hear from you.
Currently accepting inquiries for upcoming project partnerships.
Connect on LinkedIn