Andrey Dobry

Pharma BD is going global. We published benchmarks on AI for drug asset scouting with real data. Paper: arxiv.org/abs/2602.15019… #bioptic

I gave a talk at Machines Can Think on what it actually takes to apply AI in pharma. Discovery is only part of the challenge. The hard work is decision-making across science, regulation, and capital, under uncertainty. Full keynote: youtube.com/watch?v=EysO7b…

Live today on X: Machines Can Think AI Summit 2026 A full day of conversations on how AI is built, governed, and deployed at scale from national strategy to production systems. Watch live · 10:00–18:00 twitter.com/i/broadcasts/1…

Can't agree more. Reid, thanks to you and Greylock for investing in us (optic.inc) back in 2022. Since then we built a kickass AI Drug Discovery platform bioptic.io and used it to discover 4 novel drugs, including the promising small molecule for Parkinson: pubs.acs.org/doi/10.1021/ac…

We should be investing far more into AI-driven drug discovery. Every extra unit of compute has the potential to translate directly into more lives saved (and can have exponential returns). That’s the purest expression of what technology is meant to do.

I cancelled Disney+ for cancelling @jimmykimmel

🚨 Big milestone: our AI discovered new Parkinson’s drug leads. - Searched 40B molecules for $5 - 134 compounds made in 11 weeks - 14 hits, strongest at 110 nM Published today in the special issue of JCIM — my first scientific paper. pubs.acs.org/doi/10.1021/ac…

LLM-Based Agents for Competitive Landscape Mapping in Drug Asset Due Diligence 1. This paper presents a novel approach to mapping the competitive landscape for drug assets using LLM-based agents, achieving significant improvements in efficiency and accuracy for due diligence processes. 2. The system developed by the authors uses a competitor-discovery AI agent that retrieves all relevant drugs for a given indication and extracts key attributes, outperforming existing tools like OpenAI Deep Research and Perplexity Labs in terms of recall. 3. A major innovation is the introduction of a competitor validating LLM-as-a-judge agent, which filters out false positives to maximize precision and suppress hallucinations, ensuring high-quality outputs. 4. The study transforms five years of unstructured diligence memos into a structured evaluation corpus, creating a benchmark for assessing the performance of different models and agents in competitor discovery. 5. In a real-world case study with a biotech VC investment fund, the system reduced analyst turnaround time from 2.5 days to approximately 3 hours, demonstrating substantial operational impact and efficiency gains. 6. The authors compare various model configurations and show that scaffolded agents with multi-hop reasoning and web search capabilities significantly outperform single-pass prompting agents, especially on more difficult cases. 7. The system’s ability to handle multilingual, multimodal data and its robustness in dealing with fragmented and rapidly changing information landscapes highlight its potential for broader applications in the pharmaceutical industry. 8. The work underscores the importance of domain-specific benchmarks and the need for end-to-end, domain-grounded evaluation to ensure the stability and reliability of deployed AI systems in critical tasks like drug asset due diligence. 📜Paper: arxiv.org/abs/2508.16571 #LLM #DrugDiscovery #AIAgent #CompetitiveLandscape #DueDiligence #Biotech #Pharmaceuticals #Efficiency #Innovation

We just published a new paper on Teaching AI Agents to Think Like Biotech Analysts arxiv.org/abs/2508.16571 BIOPTIC agents outperform ChatGPT DeepResearch and Perplexity on real biopharma search tasks — like mapping the competitive landscape for a drug. On our benchmark, they find more true competitors than OpenAI Deep Research or Perplexity Labs. 83% recall vs 65% and 60%. They also extract structured attributes—modality, MoA, company info — clean and regulator-ready. This is how AI should work in biotech: Not just summarize text, but think like an analyst. Across science and business.

The universe is woven from energy, matter, and information. Matter — what everything is made of. Energy — what drives everything. Information — how everything is arranged. It’s all neat and logical, yet something feels missing… Love? Of course, love. Love, bits, and atoms. The rest is vacuum.

Андрей Дороничев – один из создателей мобильного YouTube и герой нашего большого выпуска про Кремниевую Долину. На момент нашей первой встречи он работал топ-менеджером Google, но вскоре ушел из компании, потеряв $10 000 000 (как именно – рассказал в этом выпуске), и создал стартап на базе искусственного интеллекта, который прямо сейчас ищет лекарство от рака. Мы встретились с Дороничевым спустя 5 лет и задали десятки стыдных вопросов про искусственный интеллект. Как именно устроен Chat GPT? Зачем Илон Маск вживляет чипы в мозг? Программисты больше не нужны? Что делать людям, которые чувствуют себя ненужными? Как обучают беспилотные авто? Как нейросеть может придумать новое лекарство? Как AI поменяет образование? Случится ли восстание машин?

The formula for success is simple: S = N × Pn, where Pn = X × R, and R = T × t. Success (S) equals the number of attempts (N) multiplied by the probability of success for each attempt (Pn). The first variable, N, is in your hands—just keep trying. The second, Pn, depends on two things: external factors (X) and your readiness (R). Your readiness R is your talent (T) multiplied by the time you’ve spent preparing (t). You fully control two variables: time spent preparing (t) and number of attempts (N). The rest is luck.

People are afraid AI will enslave them. It already happened to me. GPT is my personal trainer and nutritionist. It tells me what to eat and how to train. And I obey. Results: – 15% body fat – VO₂max 52 – 152 kg deadlift – First visible abs in my life This is the kind of tyranny I’ve been waiting for my whole life.

That unexpected and proud moment when a speaker on the opening stage of BIO 2025 says "we're actually working with an AI company called Optic".

Kicking off the #AI summit at #BIO2025 @IAmBiotech with leaders driving AI in biotech. @mmrozenman @gibsmk Don’t miss the conversation convention.bio.org/program-1/ai-s…

@denizkural @JamesSietstra

Big milestone at Bioptic! Mechanica Partners were early adopters of our generative AI for molecule design — and now they’re among the first using agenticpharma.ai. From novel drugs to strategic insight, BIOPTIC is bridging science + business. biospace.com/press-releases…

@ElliotHershberg China bio rising is scary but it’s a fact. With 1 in 3 in-licensing deals coming from China, we need to adapt. The biggest requests for the BIOPTIC platform from our customers is “can your AI agents help me discover promising assets to in-license from China?”. Yes, they can.

Awesome take by Carl June at the FDA's cell & gene therapy roundtable. Before considering trade regulations, we should look in the mirror and see what we can improve! There probably is a place for IITs in the US system.

@AndrewE_Dunn @ElliotHershberg Congratulations, @ElliotHershberg !! Excited for your new gig

Exclusive: Amplify Partners has raised $200 million for its first biotech-specific fund Targeting ~20 investments of $1.5m-$10m in pre-seed/seed/Series A-stage startups. The Silicon Valley VC has also hired @ElliotHershberg as a partner: endpoints.news/amplify-raises…

😂😂😂

Oh no… bromance is over…Who could have anticipated that 🤦‍♂️