Jike Chong

UCSF Chair of Medicine Robert Wachter (author of "The Digital Doctor" and "A Giant Leap") on Doctronic in his Substack: "Personally, I'd have no problem using this system to refill my Lipitor." He also flags that we may be the first company to buy malpractice insurance for autonomous AI in medicine. (We are.) I'm genuinely humbled by the mention. Bob has shaped how a generation of physicians thinks about technology in medicine. Everyone should read his take on why this week marked "the moment AI moves from advising on care to actually delivering it": https://lnkd.in/eqpFgC2Y

88

5 Comments

Our recent peer-reviewed paper on JMIR AI: The Leading Journal for Real-World AI Applications in Medicine with Mount Sinai Health System evaluates the performance Tailored #NLP Pipelines for Radiology, Pathology, and Progress Notes in #ClinicalDecisionSupport: https://lnkd.in/gcmFpxEK #ai #research #healthcareai #healthai #datascience #textmining

20

🚀 Just wrapped up a hands-on project applying Retrieval-Augmented Generation (RAG) to a real-world use case: interpreting LIC’s New Jeevan Shanti policy document using LLMs. The core idea? Combine vector search over regulatory PDFs with GPT-4 to generate grounded, faithful answers to complex policy questions. Key capabilities we built: ✅ LCEL-based RAG architecture using LangChain ✅ Contextual compression retriever for focused document lookup ✅ Modular LLM & embedding selection (OpenAI, HuggingFace, Mistral) ✅ Evaluation using LLM-powered metrics: faithfulness, relevance, correctness 📄 The document used is publicly available here: LIC New Jeevan Shanti Policy PDF 💻 Full code available on GitHub: https://lnkd.in/gS9mEzWA 💡 This architecture brings the best of both worlds — retrieval accuracy and generative flexibility — especially in regulated domains like insurance and financial services. 🔜 Next step: Incorporate Agents that can reason across multiple steps — not just answer a question, but decide which tools to use, when to search, how to validate, and when to stop. That’s where the future of enterprise GenAI lies. Would love to hear how others are pushing RAG, agents, and LLMs in production! #GenAI #LangChain #LLM #RAG #AIForOps #InsuranceTech #LLMEvaluation #LangChainAgents

61

1 Comment

If you are someone pushing boundaries of LLM interaction, improving prompt engineering strategies, and most importantly building more robust AI application, this paper on agentic context engineering could be a good read for you !! https://lnkd.in/gG97xZGk

39

1 Comment

𝗪𝗵𝗮𝘁 𝗱𝗼𝗲𝘀 𝗶𝘁 𝗺𝗲𝗮𝗻 𝘁𝗼 𝘀𝗰𝗮𝗹𝗲 𝗰𝗼𝗻𝘀𝘂𝗺𝗲𝗿 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝘀 𝗶𝗻 𝗮𝗻 𝗔𝗜-𝗳𝗶𝗿𝘀𝘁 𝘄𝗼𝗿𝗹𝗱? Dived deep into this at the roundtable at Bits n Atoms, hosted by The Product Folks last night — and the conversations stuck with me. Scaling consumer products in an AI world demands thinking across multiple dimensions, and these are just some of the lenses through which our group explored it.  • 𝗣𝗲𝗿𝘀𝗼𝗻𝗮���𝗶𝘀𝗮𝘁𝗶𝗼𝗻 — not just smart recommendations, but designing for one and actually impacting behaviour change  • 𝗧𝗿𝘂𝘀𝘁 — especially for this new generation of AI-native products, where UX becomes critical to adoption  • 𝗖𝗼𝘀𝘁-𝗲𝗳𝗳𝗲𝗰𝘁𝗶𝘃𝗲𝗻𝗲𝘀𝘀 — using AI to boost efficiency and also designing AI systems that can scale without exploding infra or model costs  • 𝗠𝗼𝗻𝗲𝘁𝗶𝘇𝗮𝘁𝗶𝗼𝗻 — whether through subscriptions or new models that make sense for AI-first experiences  • 𝗗𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗶𝗼𝗻 & 𝗚𝗧𝗠 — this differentiation matters more than ever, in a world where building products is getting faster and cheaper everyday  • 𝗔𝗰𝗰𝗲𝘀𝘀𝗶𝗯𝗶𝗹𝗶𝘁𝘆 — across regions, languages, devices, and modalities. If it only works for the few, does it really scale? We’re in an interesting moment — where product thinking is expanding, not just because of what AI can do, but because of the questions it’s forcing us to ask. Loved the sharp, honest, and wide-ranging perspectives from everyone Shreyanshi C. Kushal Khandelwal Ashish Jain Madhav Bhartia Tanmay Saxena Pranav Agrawal Sachin Kamkar Shashank Raghavendra Nikhil Ankan Huge shout out to Abhay Jani Aditya Mohanty Suhas Motwani for driving these thoughtfully curated and highly impactful events!

92

6 Comments

The article discusses new reranking model for question answering. Although it is designed for question answering over scientific literature, the underlying design principles are broadly applicable to other question-answering settings where semantic relevance, logical consistency, and evidential grounding are critical. https://lnkd.in/gyzQUaVa

17

A recent Stanford Medicine article highlights a clear shift in healthcare: Clinical AI has moved from theory into real-world care. The question is no longer how fast AI is being adopted. It’s how meaningfully it improves care without adding burden. For years, clinical AI stayed trapped in theory while clinicians dealt with very real documentation and workflow challenges. That gap has now closed. AI is no longer on the sidelines. It’s actively shaping how care is delivered. But here’s what matters most: Adoption isn’t the benchmark. Impact is. The industry often frames AI around task replacement. In healthcare, that framing misses the point. The real value of clinical AI shows up when it: Works with clinicians Quietly reduces friction Surfaces insight when it matters Gives time back where it counts most At OmniMD, we made an early decision. We wouldn’t build AI that adds: More clicks More alerts More noise Instead, we focus on technology that fits naturally into clinical workflows and earns trust through real-world performance — not promises. Healthcare doesn’t need more tools. It needs better ones. Because in healthcare, more is rarely better. The next chapter of clinical AI won’t be defined by hype. It will be defined by responsibility, integration, and respect for clinical expertise. Read more: https://lnkd.in/eqYmTyAP #ClinicalAI #AIinHealthcare #HealthcareInnovation #HealthTech #EHR #DigitalHealth

51

6 Comments

Even old AI models can out-diagnose human MD's. I feel like I’m choking on all of the the data. (In a good way.) A new international study pulled together cases from Harvard, Yale, Stanford, and some other academic lightweights (read: global best). Here’s the kicker—OpenAI’s o1 language model outperformed actual ER physicians on 16 out of 17 diagnostic reasoning benchmarks. We’re not just talking easy cases, either. These were “which one will you miss at 2am”-level curveballs. The o1 didn’t just keep up. It averaged 89% accuracy where the median human landed at 75%. Triage, differential diagnosis, even rare “zebras”: the model edged out real MDs. Of course, it still made mistakes (sometimes of the would-you-like-some-french-fries-with-that variety), but the stats are real. What does this mean? We’re officially crossing from “AI as a digital stethoscope” to “AI as a second (often smarter?) opinion.” The best part: every hospital, clinic, and startup founder should be sketching new workflows by Monday (or Tuesday after the holliday). “Artificial intelligence will never replace doctors. But the doctors who use it will replace those who don’t.” – E. Topol (2019) If you’re building in healthcare AI, now’s the time to move or get lapped. If you're looking to build a clinical moonshot? DM me. #HealthTech #AIinMedicine #DigitalHealth #MedEd Link: https://lnkd.in/gPjsVPr6

9

1 Comment

We are experiencing a shift in product strategy. The market is moving from model intelligence to workflow ownership. As LLMs like GPT and Claude converge on similar performance benchmarks, the model itself is becoming less of a differentiator. Intelligence is becoming a utility. The model-centric approach is fading. In this approach, the primary source of value is the behavior of the underlying foundation model itself, optimized through benchmarks and prompt engineering. The risk is constant volatility. Foundation-model updates can change behavior, alter outputs, and break assumptions the product depends on. Products built this way are often one model release away from regression, rework, or obsolescence. The workflow-centric approach is rising. It prioritizes product–workflow fit over model performance. Fit-for-purpose models are embedded deeply into vertical workflows, enriched with domain context, state, and rules. The product’s value comes from completing real tasks, which is what ultimately makes it sticky. The most successful builders/entrepreneurs I see right now are not competing on foundational architecture. They are focused on embedding intelligence directly into real workflows where work gets done. Ultimately, you can buy intelligence via an API. But there are two things you cannot buy. The first is distribution. Not sales channels, but integration. The winners optimize for presence by embedding their product deep into existing systems, delivering intelligence exactly where the work is already happening. The second is context. This is the real moat. Deep domain specialization that explains why a customer is taking an action, not just what they are typing. It is accumulated through workflow exposure and cannot be replicated by general models. In the long run, sticky workflows will beat better benchmarks.

52

3 Comments

MCP - 6 Key Questions Everyone's asking. 1. What does MCP 𝘳𝘦𝘢𝘭𝘭𝘺 mean? Remember, LLMs are just next-word predictors based on the prompt. The more context you give them along with the prompt, the better they get at "predicting." MCP stands for Model Context Protocol.  MCP is a process (Protocol) that standardizes the process for providing additional Context to LLMs through dynamic tools. The Model in MCP usually refers to a large language model (LLM), but technically, it could be any AI model. 2. Isn't that what RAG does? RAG (Retrieval-Augmented Generation) pulls data from vector stores based on user prompts and injects it into the prompt. Think about searching large documents. MCP is primarily about calling tools, including APIs, often involving real-time computation, decisions, or even user actions. RAG = stuff the model 𝘮𝘪𝘨𝘩𝘵 need MCP = ask for precisely what it needs, 𝘸𝘩𝘦𝘯 it needs it 3. Why is this suddenly a big deal? Until now, everyone has been inventing their one-off approaches: OpenAI functions, LangChain tools, and custom wrappers. MCP provides a common “USB-C”-like standard for providing context. Developers don't need to reinvent discovery, permissions, or calling mechanisms for different models. It opens up a real ecosystem of interoperable tools and AI agents. 4. Why a new protocol? What's wrong with REST APIs? REST is stateless. One call, one response. MCP is designed to support persistent sessions, bidirectional communication, and tool discovery via JSON-RPC over streaming transports like stdio or SSE. This enables a more streamlined experience. REST APIs have to be structured. MCP allows this ability to "discover" tools by the LLM based on just descriptions. MCP servers are usually implemented on top of an existing API. 5. OMG! LLM calls my APIs? Is that secure? This is a common misconception. The app (aka host) connects to the LLM, and registers available tools via an "MCP server." The LLM decides when to use a tool and sends a request back to the host. The host then calls the MCP server, which runs the tool or API. The result is sent back to the model as additional context, helping it generate a smarter response. This round trip happens in-between user interactions in the background. LLM never directly connects to the APIs or the tools. That being said, it is best practice to make sure that the MCP server is never given more permissions than is needed by the LLM. The app and the MCP server still control how, when, and what to share as context with the LLM. 6. What does this mean for healthcare? MCP combined with FHIR can fill many gaps in using AI, especially LLMs, in healthcare. We are still in the early stages of MCP evolution. Expect many new tools, especially around FHIR, to start coming up.  We at Darena Solutions | MeldRx are also exploring how to leverage this in our FHIR platform, especially around CDS. Reach out to us if you are interested in learning more.

53

21 Comments

🚀 Just published a practical introduction to LiteLLM — a lightweight but powerful tool to manage multiple LLMs across vendors. In many GenAI teams, costs spiral and wrappers multiply when you start working with many different LLM providers. LiteLLM (YC W23) provides: ✅ A unified OpenAI-compatible API framework ✅ A self-hosted proxy server with built-in monitoring, quotas, and API key management ✅ Out-of-the-box integration with LLM orchestration libraries such as LangChain and LlamaIndex. In this guide, I walk through a full setup — from Helm Charts to code — and share a few things I learned along the way.

13

When leading AI researchers validate Talbot West's original thesis. A PhD research scientist affiliated with Columbia University found some of my articles from a year ago. He reached out to share his paper, published today, that validates our claim that a modular approach is the future of AI deployment. Here's a summary of his paper: Information theory shows that pairing each AI or cognitive module with a regulator lets complex, distributed systems scale up in size and stability with only a fraction of the expected extra resource cost, opening the door to massively modular, resilient AI architectures and organizational designs. It's validating to see additional corroboration emerging, but we at Talbot West have never been in doubt about the importance of composable, modular AI architectures. That's why we developed Cognitive Hive AI (CHAI), a composable ensemble approach to complex AI deployment. #AI #modularAI #ensembleAI #composableAI #CHAI

21

5 Comments

tl;dr - I built an OpenClaw plugin for ŌURA data, and it's bonkers. One of the biggest challenges and opportunities with LLMs is to ensure that they have rich context for personalized responses. This is why OpenClaw - which has gone from zero to hero in just a few weeks - has tremendous potential (also huge risks - more on that below). If you don't know what I'm talking about, OpenClaw (formerly Clawdbot/Moltbot) is an open source project that runs on your computer and allows integration with a rapidly-growing set of tools/plugins/skills that the LLM of your choice (e.g., Anthropic, Google, xAI, OpenAI, or local models) can interact with. Imagine a personal assistant that can check-in to your flights on your behalf, install software with a simple request, or provide a summary of your day every morning. And this is all done through the messaging service of your choice: iMessage, WhatsApp, Signal, Telegram, etc. This is basically Iron Man's Jarvis, available today. I was curious how well OpenClaw would work with the high-quality, continuous data provided by Oura Ring, so I built a plugin (open source, here: https://lnkd.in/eXDZhZRs) that makes it easy for OpenClaw to query your data when it decides it would be useful. That last bit is key - modern LLMs can decide which tools would be best to answer a particular question. Disclaimer: this is a personal project and not endorsed by Oura. Use at your own risk. :) The plugin requires a few steps to set up, but it walks you through them. You can decide if you want a morning and evening summary of our Oura sent to your preferred messaging platform, and/or you can simple use it to ask questions. There have already been a few magical moments. The first was when OpenClaw proactively used my calendar data (via another plugin) to nudge me to get to sleep earlier: "💡 Sleep balance has been low for a couple days. Consider an earlier bedtime tonight — you've got swimming at 6:30 AM tomorrow." The second was when it proactively asked if it should track my sleep for a week and then follow-up afterwards (see screenshot) - real accountability! WARNING: if you don't know what a command line is, you probably shouldn't be using OpenClaw. There are real risks, including prompt injection, exfiltration of data, malicious plugins - you name it. I've set this up on a separate personal computer with local loopback and with a new/dedicated Apple Account and limited access to my actual data (including no access to my work data, and of course no connections to any AI social networks **ahem, Moltbook**!). But I learn by doing, and this is the new frontier. If you know what you're doing and want to try it out, I'd love your feedback!

157

16 Comments