Not a victory lap. Just an honest check-in on what moved, what hurt, and what’s worth doubling down on next year. This newsletter is a simple pass through my own highlights, lowlights, and funlights from 2025, and a small prompt to choose better inputs, not just chase better outcomes, in 2026.

Every biopharma leader agrees: “Pre-competitive data sharing is essential.” Yet when it’s time to act, no one moves first. Here's what I think about it- 👇

Some people think “data-centric AI” means having more data. It doesn’t. It means..... ..Having fewer regrets! #CleanData #SmarterAI #DataCentricAI

Most founders over-index on product. GTM success = knowing your ACV, product complexity, and who decides (champion or committee). “AI” isn’t a GTM differentiator, clear value is. What GTM mistake slowed you down the most? #startups #AI #healthtech

Gearing up for 𝗗𝗮𝘁𝗮𝗙𝗔𝗜𝗥 2025 in Boston. 7 years in, we ask the hard questions: • Can AI & LLMs actually speed drug discovery? • What bold bets will reset pharma? • How do we move from pilots → production? Oct 23, let’s dig in. #LifeSciences #AIinBiopharma

The "we'll build our own AI" era just ended. Takeda ↔ Nabla Bio: Renewed Sanofi: New AI deal Salt AI: $10M raised Shuttle acquires Molecule. ai The smart money figured out: AI isn't a feature you build—it's an ecosystem you join.

Earlier last week in #Philadelphia, we hosted a small group dinner during #BioTechX USA 2025. It felt great to sit down with heads of R&D, data, and operations from #pharma, #biotech, and #diagnostics hearing directly what they care about and what still trips teams up.

Clinical & real-world datasets can unlock breakthroughs in #drugdevelopment & #precisionmedicine - but patient data is fragmented across labs, imaging, omics & outcomes. Join us to see how we unify it into a #KnowledgeGraph! 👉 bit.ly/kg-webinar_p...

Data scientist: “Do we need provenance tracking?” "Yes. Because blind trust is not FDA-compliant."

One of my personal goals has always been to give back time to scientists - so they can focus on what they do best: bringing life-changing therapies to patients. This goal didn’t come from a business plan. It came from a conversation that’s stayed with me for years. 1/4

One of my biggest goals has always been simple: give scientists back time. Because behind every therapy, every discovery, is someone fighting not just biology, but time. And sometimes an extra hour can change everything. #personalstory #CEO

“A team of PhD-level experts in your pocket.” That’s how Sam Altman pitched #GPT5 We stress-tested that claim on a real #genomics problem - the kind scientists wrestle with daily in code and data. Did it live up to the hype? 👉 elucidata.io/blog/gpt5-ph... #PhD #GPT5

Delays in PK/PD data retrieval from trial PDFs can stall decisions by up to 2 weeks, push trial start dates back by months & even give competitors early access to trial sites. Time to rethink data accessibility in clinical research. 👉 bit.ly/webinar-xtract #ClinicalTrials #Discovery #AI

No one announces a “data quality failure.” It shows up as delays, poor model performance, regulatory headaches, and slipping timelines. We’re great at collecting data, less so at making it usable. In science, shaky data is a risk, not just tech debt. #Biomedicaldata #AI

I recently spoke to a biotech team scaling translational research. They have it all - EHRs, omics, imaging, trial data. On paper: perfect. In reality: chaos. Mismatched timestamps, missing metadata, systems that don’t talk. The data exists. But it’s not usable. #dataquality #AI

1 day left. Have you registered yet? 🔗 Register: bit.ly/kg_webinar #webinar #AI #knowledgegraph #research

🔗 Register: bit.ly/kg_webinar Polly KG introduces a programmable knowledge layer purpose-built for life sciences R&D. It supports natural language queries, cross-species reasoning, and domain-specific scoring within secure, enterprise environments.

After my recent post, it’s clear: the pain around knowledge graphs in biopharma is real. Disconnected schemas. No context. Engineers overloaded for basic queries. We’ve been there. We’ve felt that too. And it’s exactly why we built Polly KG.

When we built Polly KG, we weren’t trying to make just another #knowledgegraph. We built it to move past rigid schemas, surface NLP, and endless engineering loops - so scientists could just ask and answer questions that matter. Join us on Aug 5 - bit.ly/kg_webinar #webinar #AI #bioinformatics

We recently ran a closed Agentic AI workshop with biopharma R&D leaders - every seat filled. Now we’re opening it up virtually. Join us to build AI agents that actually work with biomedical data. Request your invite: bit.ly/AgenticAIwor... #AI #biopharma #clinical #workshop

Have you registered yourself yet? If not, you should do it soon. 👉 bit.ly/AgenticAIwor... #AI #data #workshops #AgenticAI

Still thinking about the energy at the Festival of Genomics & Biodata in Boston! Every R&D conversation circled back to AI in cohorting, multi-omics, and the need for better, context-rich data. Thanks to all who joined our session or stopped by. Snapshots from the talk👇

We recently ran a closed Agentic AI workshop with biopharma R&D leaders - every seat filled. Now we’re opening it up virtually. Join us to build AI agents that actually work with biomedical data. Request your invite: bit.ly/AgenticAIwor... #AI #biopharma #clinical #workshop

Join us tomorrow at 10:30 AM PT to learn what AlphaGenome delivers and how to apply its predictions effectively in your workflows. #AI #Genomics #Medicine Register: bit.ly/alphagenomew... Last day to join us!

Everyone built data lakes. Now we’re drowning in unusable data. AI doesn’t need lakes. It needs pipelines -focused, governed, built for decisions. The real bottleneck? Not compute. Not models. It’s data usability. Don’t pour more. Build better. #Dataproducts #AI #dataquality

Biomedical teams still treat data like cinema: Subscribe. Download. Adapt. Wait. But science moves like reels - live, nonlinear, always evolving. It’s not about watching the story. It’s about rewriting it - at the speed of discovery. #AIReady #DataOpsForScience #BiomedicalData #science #AI

We’re just beginning to see what foundation models like #AlphaGenome can do in functional genomics. But the real breakthroughs happen when you connect predictions to biology, context, and therapeutic intent. Join Rahul Tyagi on July 17 to see what’s next. bit.ly/alphagenomew... #AI #genomics

We’re just beginning to see what foundation models like #AlphaGenome can do in functional genomics. But the real breakthroughs happen when you connect predictions to biology, context, and therapeutic intent. Join Rahul Tyagi on July 17 to see what’s next. bit.ly/alphagenomew... #AI #genomics

Ever wondered what powers great AI? Fancy models - or clean, reliable data? In life sciences, I bet on a data-centric approach. When your data’s noisy & diverse (think omics + clinical), better data beats better models. Every time. Agree? 👇 #AI #DataQuality #BioTech #AImodel

Hallucination might be AI’s most human trait. Science evolves through messy, imperfect first drafts. LLMs aren’t built for truth - they’re built for possibility. Use them not to answer, but to explore. That’s where discovery begins. #AI #LLM #Science #Discovery #AgenticAI

🇺🇸 Happy Independence Day from all of us at Elucidata! To our colleagues, customers, advisors, and partners in the U.S., thank you for your vision, conviction, and collaboration. Cheers to a peaceful, safe, and happy holiday. #IndependenceDay #4thofJuly #Happy4th #July4th

Built on GPT-4.1 and powered by tree-of-thought reasoning, our new multi-agent AI system outperforms standard prompting baselines across critical metadata fields like disease, tissue, sampling site, donor mapping, demographics, and more. Here’s what it delivers:

Stop believing "not enough data" is a blocker. It's not. Quality data wins over quantity every time. A small, high-quality dataset is potent. Focus on precision, not just volume. #leadership #DataScience #AI

One of the most common traps early founders fall into is .. Focusing on what's easy to answer, not what’s essential to decide. Should I use HubSpot or Notion? Should I set up a Delaware C-Corp now or later? SAFE or convertible note? Important? Sure. Defining? No.

Just wrapped a workshop in Boston where we built Agentic #AI systems from scratch. What breaks most projects? Messy data. What makes them work? Structure, context & scale -starting at the data layer. More soon, including a deep dive on our system: bit.ly/AgenticAIweb...

Most scientist-founders aren’t stuck because they filed the wrong paperwork. They’re stuck because they never asked: Am I building a 5-year platform or a 3-month product? Scientific clarity ≠ business clarity. Ask the hard architectural questions - early 😃 #startups #founders

Looking forward to speaking at #FOGBoston today! 😃 Let me know if you are here and want to chat! Exciting times for #genomics data 🧬 @flgenomics.bsky.social

High-quality metadata is essential for downstream AI applications; however, it remains the weakest link in most biomedical pipelines. In our latest preprint, we introduced a novel Multi-Agentic AI System built to break the tradeoff between metadata quality and scalability:

#Genomic data complexity slowing down your oncology, biomarker, or clinical trial work? We get it. That's why we're excited for the Festival of Genomics & Biodata Boston 2025! Come chat with us about tackling disjointed multi-omics data and accelerating insights. @flgenomics.bsky.social

Automated metadata curation with human-level accuracy, dataset-wide scale, and traceable quality. Our Agentic AI system extracted 23+ metadata fields with 93% recall from GEO, papers, and supplementary files across 211 studies.

Last night’s dinner in Boston reminded me why we do this :) A few R&D leaders around a table, sharing what’s working, what’s not, and how to make biomedical research actually move at the speed we need.

AI agents fail when built on fragmented, context-agnostic data. In discovery, translational and clinical, that failure costs time, capital, and patient outcomes. On June 20, we are hosting an in-person workshop in Boston on Building AI Agents with Fit-for-Purpose Data.

📣 We’re excited to share our latest work! A first-of-its-kind multi-agent AI system for high-quality, scalable metadata curation in biomedical datasets. Yes, we’re talking about those public datasets (👀 GEO) where the data is gold, but the metadata is a mess. 👇 www.biorxiv.org/content/10.1...

Join us at the BIO International Convention 2025. I’m meeting teams rethinking how data infrastructure powers modern drug discovery. 👉 www.elucidata.io/event/bio-in...

I am thrilled to announce our partnership with Sapien Biosciences, India’s largest commercial biobank with over 300,000 patient samples. Read more: www.elucidata.io/company/indu...

This is happening today. Last chance to register!

The U.S. still sets the bar when it comes to scientific discovery. There’s no denying the strength of its academic institutions, the depth of its clinical research infrastructure, or the caliber of its scientists.

The industry has been obsessed with speed. But what if the better question is: speed to what? We’re seeing a shift in how early-stage science gets done: Companies using patient samples earlier in the pipeline Biotechs bypassing traditional preclinical steps

Most teams building AI into biotech don’t need the “best” model. They need the right one for the job. Two real tasks we see: Summarize 80 papers into a cited report → Claude shines. Annotate microscopy images with metadata → Gemini fits. The smart strategy? 👇

Ever seen cell-type maps generated from just H&E slides - no spatial data needed? The future is here. ML scientists are decoding spatial biology with AI + histology alone. Join us at our Virtual Symposium on June 7 | 11:30 - 1:30 PM EDT 👉 t.co/6Fs2WVvRJJ