Zak Doric, PhD

Protected by 60 patents covering hardware, software, data, and AI models, with science that is 15 years in the making - Apoha is finally coming out of stealth 🥷 Alongside a Fortune exclusive, the team is releasing two scientific preprints making the underlying science publicly available for the scientific community to review & build on. 𝗕𝗮𝗰𝗸𝗲𝗱 𝗯𝘆 $𝟯𝟲𝗠 𝗶𝗻 𝗳𝘂𝗻𝗱𝗶𝗻𝗴 from Singular, redalpine, Draper Associates, Seedcamp, Acequia Capital, Wilbe and Nucleus Capital, the team is building one of the most consequential technology platforms of this cycle with applications across Pharma, Food, Materials & Physical-world AI. Congratulations to the entire team, investors and customers for embarking on this mind-blowing journey! 🥂 Anshika Srivastava, Shamit Shrivastava, Akshay Mishra, Carolina Zirn #venturecapital #deeptech #techbio

74

4 Comments

EXCLUSIVE: Three young MIT researchers are launching their new biotech startup today, having raised a $28 million seed round from investors like a16z, Amplify Partners, and Zetta Venture Partners. The company, called Boltz, was formed in late 2024, partially in response to seeing AI models in biology growing more capable yet less accessible. As a public-benefit corporation, Boltz aims to keep open-sourcing its major models, while making a business in selling software and services in deploying these in the lab. “What we started seeing with AlphaFold 3 is the best models are becoming more and more closed,” CEO Gabriele Corso told me. “The level of impact of these models is how good the models are multiplied by the number of people who are able to access them. We want to maximize both sides of the equation.” More on the future of a 13-employee startup worth watching closely in Endpoints News: https://lnkd.in/eTiFrQy8

546

14 Comments

Biology has spent its entire history as read-only. The read-write era starts now. Radical Numerics, the pioneers of generative genomics, is writing original, functional biology from scratch using AI. We are thrilled to announce that Emergence Capital is leading their $50M seed round, alongside Obvious Ventures, Triatomic Capital, First Spark Ventures, and Factory. Instead of narrow, single-purpose AI, they are building a general biological intelligence that learns across DNA, RNA, proteins, and more simultaneously. Crucially, they operate under a strict dual mandate: advance biological design to transform medicine, and build the critical biodefenses to protect it. Welcome to the Emergence family, Eric Nguyen, Michael Poli, and the full Radical Numerics team. Read more from Gordon here: https://lnkd.in/gSBu6Bfy

47

1 Comment

Anthropic just launched Claude for #Healthcare, and OpenAI just launched ChatGPT Health. Both are LLMs, but their strategies differ significantly. I wrote down a few thoughts. 1️⃣ 𝐂𝐥𝐚𝐮𝐝𝐞 𝐰𝐚𝐧𝐭𝐬 𝐭𝐨 𝐛𝐞𝐜𝐨𝐦𝐞 𝐭𝐡𝐞 𝐧𝐞𝐰 𝐨𝐩𝐞𝐫𝐚𝐭𝐢𝐧𝐠 𝐬𝐲𝐬𝐭𝐞𝐦 𝐨𝐟 𝐡𝐞𝐚𝐥𝐭𝐡𝐜𝐚𝐫𝐞 Claude is entering the race with the ambition to integrate AI scribe + AI billing + AI clinical decision support + workflow, in order to become a new intelligence layer on top of legacy systems. It targets payors, providers, and consumers, tapping into (public) databases to optimize clinical operations. Claude also launched Claude for Life Sciences in October 2025, offering tailored solutions for pharma and integrating them into this broader ecosystem. 2️⃣ 𝐂𝐡𝐚𝐭𝐆𝐏𝐓 𝐇𝐞𝐚𝐥𝐭𝐡: 𝐭𝐡𝐞 𝐟𝐮𝐥𝐥 𝐜𝐨𝐧𝐬𝐮𝐦𝐞𝐫 / 𝐩𝐚𝐭𝐢𝐞𝐧𝐭 𝐩𝐥𝐚𝐲 ChatGPT’s strategy is much narrower, as it focuses primarily on consumers and patients, but with a much deeper offering. It integrates medical navigation, lifestyle, nutrition, and fitness, as well as decision support to help users choose the right care plan (in the US), going after players like Transcarent. 3️⃣ 𝐁𝐨𝐭𝐡 𝐂𝐥𝐚𝐮𝐝𝐞 𝐚𝐧𝐝 𝐂𝐡𝐚𝐭𝐆𝐏𝐓 𝐩𝐫𝐨𝐯𝐢𝐝𝐞 𝐭𝐡𝐞 𝐢𝐧𝐟𝐫𝐚𝐬𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞 𝐟𝐨𝐫 𝐜𝐨𝐧𝐬𝐮𝐦𝐞𝐫 𝐀𝐈 Both Claude and ChatGPT integrate with Function Health, which uses blood analyses to both detect diseases early and optimize daily health. Consumer AI currently remains a tool for the healthy and wealthy, but over time the system may shift from being doctor-centric to consumer-centric. Claude and ChatGPT are clearly accelerating this shift. 4️⃣ 𝐖𝐡𝐨 𝐰𝐢𝐥𝐥 𝐰𝐢𝐧: 𝐂𝐥𝐚𝐮𝐝𝐞’𝐬 𝐢𝐧𝐟𝐫𝐚𝐬𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞 𝐩𝐥𝐚𝐲 𝐨𝐫 𝐂𝐡𝐚𝐭𝐆𝐏𝐓’𝐬 𝐜𝐨𝐧𝐬𝐮𝐦𝐞𝐫 𝐡𝐞𝐚𝐥𝐭𝐡 𝐛𝐞𝐭? Claude is trying to go very big: maybe too big. No one really wants a single player to own such a large and strategic piece of the healthcare tech stack. ChatGPT, on the other hand, aims to become the digital front door of healthcare, similar to Apple or Amazon. In the short term, it may monetize through advertising (as OpenAI has recently launched ads on its models). In the long term, it’s unclear how it will evolve beyond this. It is already replacing “Dr. Google”, but Google itself never managed to build a truly large business around that. 5️⃣ 𝐂𝐥𝐚𝐮𝐝𝐞 𝐢𝐬 𝐫𝐞𝐝𝐞𝐟𝐢𝐧𝐢𝐧𝐠 𝐢𝐧𝐭𝐞𝐫𝐨𝐩𝐞𝐫𝐚𝐛𝐢𝐥𝐢𝐭𝐲 Claude connects to many different sources, often involving unstructured data. Thanks to LLMs, it does not require highly structured or standardized data to operate effectively. MCPs combined with LLMs are redefining interoperability. Claude can ingest raw, unstructured data from multiple systems and generate usable insights. Traditional IT and SaaS systems were simply not able to do this before. ---- Follow HGM Advisory for daily reposts of the most interesting #HealthTech news.

794

71 Comments

One of our most-read stories of 2025: A Stanford study tested five popular therapy chatbots and discovered they stigmatize conditions like schizophrenia and alcohol dependence. AI might excel at paperwork, but healing humans still requires other humans in the room. Read the full study: https://lnkd.in/gJ3Fg_RX

97

1 Comment

The latest by Insight Innovation Ventures (IIV): Not All Evidence Gets a Vote AI can generate more evidence than ever. The next control point is deciding what evidence is allowed to authorize which decisions. In Where Value Compounds: A Field Guide for the Decision Economy, we argued that two questions determine where value actually accumulates as AI restructures the insights and analytics industry: How consequential is the decision this offering feeds? How replaceable is the input itself? Cross those two variables, put trust in front of them as a gate rather than a dial, and most of the industry becomes easier to read. Low-consequence, highly substitutable work will automate and commoditize. High-consequence work supported by scarce, trusted evidence becomes decision infrastructure. Workflow embedding can reinforce either position, but it cannot turn a replaceable input into an irreplaceable one. That piece ended with a promise. The next step was to build the instrument behind the map rather than stop at a useful way of looking. This is the beginning of that next step. The instrument is an Evidence Admissibility Matrix (EAM): a system for determining which kinds of evidence are allowed to support which kinds of decisions, under what conditions, with what level of human review, and with what authority to act. The canonical beta specification is complete: decision-authority ladder, evidence classes, assurance gates, agent-permission rules, machine-readable schemas, and an evidence-passport contract. Detailed scoring weights, exception logic, and operating thresholds remain proprietary. The framework is in beta, not v1: independent pressure-testing comes next. Want to pressure-test the beta on a real claim–decision workflow? Email lmurphy@greenbook.org Full piece: https://lnkd.in/e54M-6Fj Thanks to Elaine Riddell, Rolfe William Swinton, Gregg Archibald, Shane Skillen and many others for all the work in helping with inspiration, the build and early testing!

11

2 Comments

We've co-led Fusionality's CHF 3m pre-seed alongside our friends at Founderful Every fusion device has to solve the same problem. Measure the plasma, infer its state, act on heating and magnets, all in milliseconds, before it gets away from you Federico Felici and Jonas Buchli met while teaching an AI to control EPFL's TCV tokamak, work published in Nature in 2022. Jonas was at Google DeepMind at the time and Federico joined afterwards, building simulations and ML interfaces for fusion devices. They have spent years solving exactly the problem their customers now have. Fusionality is the integrated hardware and software layer for measurement, simulation and control, built to move between devices with minimal customisation so each deployment costs meaningfully less than the last. Why this matters now: 56 private fusion companies raised $4.48bn last year, and almost all of them are rebuilding the same control loop from scratch. Around 80% of any given control system is common across designs. Fusion is scaling faster than its supply chain, and operations infrastructure is the clearest gap in it. Five companies have signed offtake agreements, six have siting agreements. These are becoming contracted power plants rather than experiments. Congratulations Federico, Jonas and the whole Fusionality team, and thank you for the trust. Full thesis in the comments below https://lnkd.in/eM9Sidg3

49

4 Comments

Rather than hallucination, sycophancy in LLMs (Generative AI) is why its use in healthcare is so much more dangerous than we currently understand. We are currently conducting a massive social experiment with almost no clinical safety guardrails. https://lnkd.in/gfHWTCDV #digitalhealth #ai #llm Dr John F. Lambert Anja Nikolic Kate Ebrill Michael Lawley John Grimes Joanne Lee Alyce Cuman Liam Cahill Jay Rebbeck

47

16 Comments

AI in life sciences is starting to feel less like a headline and more like a quiet revolution. Claude has just realised a dedicated AI model for life sciences.  Researchers can now:•  -Parse and summarise entire libraries of literature in hours, not weeks. -Draft study protocols, consent forms, and SOPs with regulatory precision. -Translate raw clinical and genomic data into readable insights. In a field where accuracy and compliance matter as much as speed, Claude is one of the few systems built for scientific rigour, not marketing noise. Worth checking out if you’re operating anywhere between discovery and delivery: https://lnkd.in/eaFZjd3t

11