AI product leader / drug discovery / agentic AI

MassimoSammito.

I turn frontier AI into drug-discovery products that scientists adopt.

I set the AI product strategy

Associate Director, Product at Recursion · Cambridge, UK

10+years across structural biology, drug discovery & AI products
200+scientists on the platform I owned at AstraZeneca
30%less antibody design time from GenAI workflows
25+peer-reviewed papers · h-index 21
01

Approach

A scientist who builds products. Not a lab, not a black box.

I started by writing crystallographic phasing algorithms, then designed antibodies with AI, then led the product organisations that put AI in front of scientists. That path is the job: understanding the science well enough to know what matters, and the AI well enough to know what is real.

A

Set the strategy

Vision and roadmaps for AI portfolios — today an integrated agentic AI ecosystem across the design-make-test-learn cycle at Recursion; before that the Biologics Discovery Platform at AstraZeneca.

B

Translate across teams

Turning biological needs into technical roadmaps, and model results into decisions — between wet-lab and computational teams, AI research and discovery, science and the business.

C

Prototype to de-risk

I build working prototypes with Claude Code and Codex — spec first, then agents — so an idea becomes something concrete before a team commits to it.

D

Lead product teams

A ~20-person organisation with 8 direct reports at AstraZeneca; a team of 4 Product Managers at Recursion. Science-first, agile, accountable for adoption.

02

Experience

From phasing algorithms to agentic products.

  1. Mar 2026 — now

    Associate Director, Product

    Recursion · UK

    • Product leadership — lead a team of 4 Product Managers, each owning a portfolio of AI and agentic products across the drug-discovery pipeline.
    • Agentic DMTL ecosystem — set the vision and roadmap for an integrated agentic AI ecosystem across the design-make-test-learn cycle; own the Design layer and lead the implementation of Test and Learn.
    • Hit-ID & Design — own the Hit Identification & Design product area, with 2–3 agentic AI products in pilot with computational scientists.
    • Research to product — partner with Valence Labs, Recursion's AI research unit, to bring models in transcriptomics, genomics and de novo design into products for discovery teams.
    Agentic AIPortfolio roadmapHit IDTeam of 4 PMs
  2. Oct 2024 — Jan 2026

    Director, R&D IT — Data, Analytics & AI

    AstraZeneca · Cambridge, UK

    • Platform ownership — product vision and roadmap for the Biologics Discovery Platform, serving 200+ scientists globally.
    • Shipped AI products — a developability-prediction product that won the internal SVP Award, and GenAI workflows that cut overall antibody design time by 30%.
    • Team leadership — a ~20-person product organisation with 8 direct reports: Product Owners, AI Scientists and Engineers.
    • AI governance — set up and chaired the AI risk-governance framework; 100% of deployed models and data products met internal regulatory and safety standards.
    • Vendors & ecosystem — moved Seqera (Nextflow) to usage-based licensing; 15% efficiency gain with Schrödinger on token optimisation; collaborations with NVIDIA (model fine-tuning) and Rhino Health (federated learning).
    • Scientific velocity — on-demand AWS EC2 sandboxes removed the screening-campaign bottleneck and sped up scientific delivery by 50%.
    SVP Award200+ usersAI governance8 direct reports
  3. Aug 2022 — Oct 2024

    Associate Principal Scientist — Strategic AI Lead, Antibody Discovery

    AstraZeneca · Cambridge, UK

    • AI-designed antibodies — led AI-driven discovery for 3+ campaigns that produced experimentally validated, high-affinity lead candidates.
    • Hybrid modelling — drove the shift to hybrid physics-based + deep-learning workflows, improving prediction accuracy by 15% over legacy methods.
    • Data strategy — curated 1.4M antibody sequences (OAS) and 7k structures (SAbDab) into a proprietary training asset.
    • Alliances — managed partnerships with AI biotechs Nabla Bio and Absci across 4 joint discovery targets.
    • Bridge role — the liaison between wet-lab and computational teams, turning biological needs into executable technical roadmaps.
    Antibody designNabla Bio · AbsciData strategy
  4. Jun 2021 — Aug 2022

    Senior Computational Scientist II

    PhoreMost · Cambridge, UK

    • Designed AI/ML pipelines for peptide-library engineering that fed directly into R&D decisions; replaced manual analysis with automated workflows and brought agile practice to the science team.
  5. May 2018 — Jun 2021

    Marie Skłodowska-Curie Individual Fellow

    University of Cambridge · Cambridge, UK

    • Project leadership — secured competitive EU funding and managed the end-to-end lifecycle of scientific software development, delivering tools for X-ray crystallographic data phasing.
    • Impact — developed reproducible methodologies for structural biology, published in high-impact journals, including early validation of AlphaFold / CASP14 models for structure solution.
    EU fellowshipAlphaFold · CASP14
  6. Jan 2017 — May 2018

    Postdoctoral Fellow

    C.S.I.C. · Spain

    • Developed computational methods for X-ray crystallography and data clustering; supported research delivery and reproducibility.
    • Supervised and mentored PhD students in research projects.
  7. Sep 2015 — May 2016

    Postdoctoral Researcher

    Georg-August-Universität Göttingen · Germany

    • Computational research in structural biology focused on algorithm development.
  8. Jan 2011 — Sep 2015

    Bioinformatics Computer Scientist (PhD student)

    C.S.I.C. · Spain

    • Developed computational algorithms for crystallography and applied machine learning to structural bioinformatics.
    • Published peer-reviewed work in computational biology and structure determination, including a first-author Nature Methods paper.
    ARCIMBOLDONature Methods
03

Lab

I still build. It keeps the strategy honest.

In my free time I engineer scientific and educational apps for my own use with Claude and Codex. It is how I keep up with what the tools can really do.

  1. Design docWhat problem, for whom, how we'll know.
  2. Technical specData, interfaces, failure modes.
  3. Agents buildClaude Code · Codex.
  4. A working prototypeSomething real to react to.

Agentic prototyping

Functional prototypes with Claude Code and Codex using a spec-first workflow, plus a local multimodal model stack.

Architecture literacy

Agentic architectures, multi-agent workflows and software-engineering practice — enough to shape technical direction with engineering and research teams.

Commit activity · last 6 months Private repositories · counts only · updated 19 Nov 2026
Commits
1,155
Active days
58
Longest streak
14 days
05

Background

Education

PhD, BiotechnologyComputational Crystallography · Universitat de Barcelona2015
MSc, BioinformaticsDual degree · Universitat Pompeu Fabra & Università di Bologna2011
BSc, Computer ScienceUniversità di Catania2009

Toolkit

AI & science

LLM & agentic products · protein & antibody design models · developability prediction · federated learning · model evaluation

Product

Portfolio prioritisation · roadmapping · product discovery · OKRs · Agile · Airfocus · JIRA · Confluence · Smartsheet · Miro

Technical

Python · PyTorch · SQL · Java · C++ · AWS (Bedrock, S3, EC2) · Kubernetes · Nextflow · Claude Code · Codex · Cursor · Gemini

Languages

Italian (native) · English (full professional) · Spanish (full professional) · French (intermediate)

06

Contact

Let's talk about AI that scientists actually use.

AI product strategy in drug discovery, agentic systems for science, or structural biology.