Opinionated and unfiltered thoughts of an experienced full stack dev are loading…
Haven't posted in 1yr! Amazing convos with builders at Miami Consensus & Solana Accelerate energized me to start building out in the open.
@OpenMed_AI@nxpatel@EvidenceOpen The agent has an evidence layer built in?! How can I get involved as a dev? Would love to build an interface so my wife can test it at her clinic.
Looks promising, just missing @EvidenceOpen. Is there an open-source alternative, or can the evidence layer even be open source? Pharma pays for ad placement mid-query to clinicians in OpenEvidence, which is why I'd avoid building on it as a developer. Not saying that biases the answers, but an evidence layer that doesn't depend on this ad model seems essential for "Forward Deployed Clinicians" to innovate without being tied to a pharma incentive structure.
@MarioATX_MD Love it. You're right about maintenance/exposure, cloud seems the obvious choice. Vercel won by making deploys trivial, whoever makes AI painless/safe for clinicians to deploy wins. Let me know when I can approach a clinician's office and say "Own your Office + AI" via Cline SDK.
@Hihihqi3@nxpatel Does the FDP need to live inside a closed frontier lab, or could the same person have more leverage building in the open? What's pulling you toward the lab side?
@Bencera Honest feedback: When I tried Polsia, it felt like classic Ai slop wrapped in dark patterns. If that wasn’t your goal, I hope you spend funds on closing the gaps on your products claims, instead of just 'tokens + marketing'. Best of luck and keep building!
@GoogleDeepMind I upgraded to ultra in @FlowbyGoogle and this is my experience with Omni Flash... and the Story board tool or Agent are a painful experience. Is anyone actually getting good results with this?! HOW?
@itsEmZee_@HealthcareAIGuy Why do we even need Epic or OpenEvidence at all?Agent-native (100% local) is obviously the future. Just pull in whatever evidence/guidelines the agent needs from the same sources they use. Clinician should be able to finish a patient workflow without needing to jump hoops.
It's a good hill to die on. Clinicians can already achieve that individualized workflow today with private local LLMs combined with open-source EHR/EMR solutions. It's not obvious to most because it's not profitable for established players to admit this, and it's not easy for clinicians who prefer focusing on patient care to build their own agents when it's not the standard. This represents a fundamental shift in clinical workflows, and I expect you'll be proven right with this approach very soon, with the rise of more physician-founded Telehealth and health startups.
This is amazing and the only reason I can think of that it’s not going viral, is that you’re early. Once people start to realize they don’t have to go through huge vendors anymore to make software in this industry, it will be disruptive, to say the least! Following closely and building something adjacent to this space.
@trq212 I've been trying this pattern in different ways but still failing when complexity is high. Wish there was something similar to pre-commit hooks in claude. Even if I spent a lot of time in spec/planning, it always drifts during execution... feels like something is missing here.