Nebojsa V.

Langgraphs and multi agentic flow with contextual memory can enable the best customer service. We see a future where the Lang graphs deep technology over open source platforms can enable data structuring and even a better informed ai assistants throughout the spectrum. #langchain #crewai #llm #langgraph #andrewng DocsTribe AI

3

OpenAI just announced.......... No, they didn't. I did. Introducing the latest update of my startup ChatDoc. ChatDoc Alpha. ---> Just came up with that. Hi, I'm Joshua Anang, I'm 17 and I'm the Founder and CEO of ChatDoc. At ChatDoc, we're providing a virtual healthcare assistant. To engage in interactive, mental and medical conversations. Our main goal is to: Build a world where everyone has a doctor. Overall, ChatDoc has a variety of use cases. ChatDoc can be used for: - Symptom checking - Early disease detection - Healthcare assistance - Interactive healthcare conversations. Our target market is the United States. We've been able to hit over 200 users on our MVP since launch. We're reaching out to investors to scale our operations and make an impact. Making healthcare portable and accessible will immensely improve the quality of life. And with ChatDoc at the helm, what's impossible is only the impossibility. Dominique Kim, MPH Suzy Engwall Uli K. Chettipally, MD., MPH. Nila Rafiq, MD, MBA Harvey Castro, MD, MBA. Kelston S. Alister Martin Dylan Attard MD, MRCSI, MEnt. Dr. Achim Simons Nirmal Patel Hannah Cooper Klein Ela Bergaoui Ivan Tornos Rafael M Lopes, MBA, CPHIMS, CPXP Dipu Patel, DMSc, MPAS, ABAIM, PA-C 🚀 Mercedes Bankston Flora Wan, MASc, PMP Robert Hoyt MD FACP FAMIA ABPM-CI Sean Tilson Seth DeHart Matt A. Joshua Y. Lee Eve Cunningham MD MBA Rebecca Thomas Carolina Batista MD Feham Riaz Gary Goldman MD, DDS, MBA Irina Kukuyeva, Ph.D. (she/her) Jan Beger Tom Navasero Vijay K. Luthra Konrad Komnata Aleksandrina Ikonomova - Allya Sigrid Berge van Rooijen João Bocas Mujahid Al Sayed 🏁Guc Ozenci #healthcare #healthtech #ai #healthcareai #digitalhealth #digitalhealthcare #startup #healthcareimprovement #healthcareinnovation #healthcaretechnology #medtech #healthai #AIinHealthcare #HealthTech #globalhealth #healthcareequity #HealthEquity #GlobalHealthEquity #GlobalHealth #artificialintelligence #ai #chatbot #virtualassistant #virtualcare #virtualhealthcare #openai #technology #GPT #AGI #innovation #MachineLearning #LLM #MultimodalLLMs #deeplearning #investments #investor #venturecapital #VC #VentureCapitalist

15

12 Comments

💡64% of outcomes come from just 4% of causes This insight comes from recursively applying the Pareto Principle (or the 80/20 rule) which states that roughly 80% of outcomes come from 20% of causes. This can be seen in many different areas and disciplines - 20% of patients incur 80% of healthcare expenses - 80% of a company’s profits come from 20% of customers At Cruise, we employed this principle through continuous application of Pareto analysis, helping us work effectively towards improving the safety of the car. The idea is to rank occurrences of problems by frequency, and focus on the highest ranking issues first. By addressing just 20% of the top causes, you can expect roughly 80% of problems to be addressed. Additionally, by continuously applying this principle in a cycle, addressing the most impactful causes each time, you can achieve a form of exponential improvement, as each cycle’s improvements build upon those of the previous one. So when you’re overwhelmed by tasks and demands, just remember that if you focus on just 4% of them, you can achieve 64% of the results! With a little prioritisation and focus, you can significantly improve your output 📈

21

2 Comments

ENCHIRIDION Devlog #1: A 2.5D environment, but why? The Enchiridion will be a mobile health improvement game based on the story of Epictetus the philosopher. Set in a near future you will play as Ari, a somewhat lost individual that illegaly grows and sells vegetables. Grow vegetables to earn money, improve the community you live in, do workouts with Ari to maintain his mental balance, meditate to improve his focus and learn about the basic philosophy enclosed in the ancient book "The Enchiridion". Follow me on https://lnkd.in/eKQfFrwd for all the updates! #gamedev #workout #games #mobilegaming

10

First test email from Zeeq.ai came through overnight. Zeeq (rhymes with "seek") is a personalized AI agent that watches ClinicalTrials.gov for updates of #clinicaltrials for trials of interest and generates answers to your questions delivered straight to your inbox whenever a trial gets added or updated. Had a few hiccups with reading the change feed from the FDA, but looks like it's almost there and getting closer to full rollout (a few more days of observing, I think!). Whether you are a #patient, #pharma exec, #biotech industry analyst, or academic researcher, Zeeq can help you stay on top of the most relevant trials of interest with personalized questions and answers. Free to sign up at https://zeeq.ai - no usernames or passwords, no accounts.

11

3 Comments

Early-stage fundraising is getting complicated due to SAFEs. Should they be limited to early rounds? • Angel Rounds (SAFEs): $150K-$200K raised, small valuation caps, light dilution. • Small Pre-Seed (SAFEs): $680K raised, $9M cap, 6%-8% dilution. • Big Pre-Seed (SAFEs): $1.5M raised, $12M cap, 11%-13% dilution. • Seed (Priced vs SAFEs): ~$3.3M raised, $13.7M valuation (priced) vs $18M cap (SAFEs), slightly lower dilution with SAFEs. • Post-Seed Extensions: More common, but costly in equity. #Fundraising #Startups #VC #Investment

7

Ever get frustrated by how fragile prompt engineering can be? Or wish there was a more systematic way to build language model pipelines without countless rounds of trial and error? Meet DSPy – a new way to think about AI. Instead of painstakingly crafting prompt templates, DSPy allows us to build self-optimizing AI pipelines using modular blocks. Imagine turning complex LM operations into simple, reusable functions that learn to improve over time. 𝑯𝒆𝒓𝒆’𝒔 𝒘𝒉𝒚 𝑫𝑺𝑷𝒚 𝒊𝒔 𝒂 𝑮𝒂𝒎𝒆-𝑪𝒉𝒂𝒏𝒈𝒆𝒓 𝒇𝒐𝒓 𝒔𝒐𝒇𝒕𝒘𝒂𝒓𝒆 𝒅𝒆𝒗𝒆𝒍𝒐𝒑𝒆𝒓𝒔:- 𝟭.⁠ ⁠𝗠𝗼𝗱𝘂𝗹𝗮𝗿 𝗔𝗜 𝗦𝘆𝘀𝘁𝗲𝗺𝘀: DSPy replaces brittle prompt templates with reusable modules like Predict, ChainOfThought, and ReAct. These modules generalize popular prompting techniques and can be optimized automatically, much like neural network layers but for LMs. 𝟮.⁠ ⁠𝗦𝗲𝗹𝗳-𝗢𝗽𝘁𝗶𝗺𝗶𝘇𝗶𝗻𝗴 𝗣𝗶𝗽𝗲𝗹𝗶𝗻𝗲𝘀: DSPy has a built-in compiler that optimizes the pipeline by bootstrapping demonstrations. This means the system can learn to improve its modules autonomously, much like training a neural network, but focused on refining how LMs are used in a pipeline. 𝟯.⁠ ⁠𝗘𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝗰𝘆 𝗚𝗮𝗶𝗻𝘀 𝘄𝗶𝘁𝗵 𝗦𝗺𝗮𝗹𝗹𝗲𝗿 𝗠𝗼𝗱𝗲𝗹𝘀: DSPy allows smaller models like Llama2-13b-chat to achieve performance competitive with larger models by optimizing the pipeline design. Case studies show DSPy outperforming standard few-shot prompting by 25%-65%, and expert-crafted pipelines by up to 46%. 𝟰.⁠ ⁠𝗗𝗲𝗰𝗹𝗮𝗿𝗮𝘁𝗶𝘃𝗲 𝗣𝗿𝗼𝗴𝗿𝗮𝗺𝗺𝗶𝗻𝗴 𝗠𝗼𝗱𝗲𝗹: DSPy uses natural language signatures like question -> answer to abstract LM operations. Modules like ChainOfThought can be easily plugged in, making it simple to build complex workflows involving multiple LMs, retrieval steps, and reasoning processes. 𝟱.⁠ ⁠𝗖𝗼𝗺𝗽𝗶𝗹𝗲𝗿-𝗗𝗿𝗶𝘃𝗲𝗻 𝗢𝗽𝘁𝗶𝗺𝗶𝘀𝗮𝘁𝗶𝗼𝗻: DSPy’s compiler simulates the pipeline on training inputs, optimising prompts, instructions, and LM selection. This systematic approach reduces the trial-and-error of prompt engineering, making pipelines more adaptable and scalable. Whether you’re a techie knee-deep in LLMs or just curious about the future of AI, DSPy is a paradigm shift. It’s not just about building better models but about creating smarter processes for AI development. Check out the paper here - https://lnkd.in/gKeizd8J #SaaS #AI #Tech #Engineer #Retail

21

I'm trying to integrate Redis to manage real-time messaging between patients and doctors. 📝 Here's a quick snapshot of what I’ve implemented: 📌Subscription to User Channels: Utilizing Redis' Pub/Sub mechanism, users can subscribe to channels dynamically based on their IDs. This ensures they receive messages directed specifically to them in real-time. 📌Message Broadcasting: When a user sends a message, it's published to the Redis channel, allowing the target user to receive it instantly, no matter where they are connected from. Looking forward to continuing this journey and sharing more updates soon! 💻🎉 #fullstackdeveloper #jobhunting #TechForGood #TypeScript #Prisma #TailwindCSS #TurboRepo #FullStackDevelopment

4

Large Language Models (LLMs) as super lossy compression. I have been thinking a lot about demystifying what LLMs do because lots of y'all's are very mystified. In the linked article, I try to reframe them for you, hoping it will help you understand what they good for and what they are not good for. Side note: I am still looking for my next adventure. If communicating complicated concepts so that regular people can grasp them is something you need, you should message me. I write code and stuff too and am usually fun to be around. #WrittenByMe

6

3 Comments

Large codebases are scary! 😱 When you're starting out contributing to a mammoth codebase like SigNoz with over 300000 lines of code, you have no idea where to even start looking. We took an open issue from Signoz's repository and asked Potpie AI's low level design agent to help us get started. The results were mindblowing 🤯. For a repository with THOUSANDS of files, Potpie was able to : * Pinpoint which exact file needed to be updated * Tell which part of the code needed updating * Suggest next steps like testing 🥧 agents >> Try us out at https://potpie.ai We are open source! Pranay Prateek Ankit Nayan Shivanshu Raj Shrivastava

51

5 Comments

💣 Google just dropped a bomb with Alphafold 3 - the biggest leap forward in AI for biology since Alphafold 2 - extending Alphafold 2 capacity of predicting proteins structure to DNA, RNA, small molecules, ions, ligands and their interactions ! Alphafold 2 - Earlier versions focused on predicting the 3D structures of 200m proteins from their chemical constituents. Knowing what shape a protein takes is crucial because it determines how the protein will function – or malfunction – inside a living organism. Alphafold 3 - Today in Nature Demis Hassabis, CEO of Deepmind, describes how AlphaFold3 can predict how proteins interact with other proteins, ions, strands of genetic code, and smaller molecules, such as those developed for medicines. Co-developped with big pharma Novartis and Eli Lilly. Performance: Alphafold 3 versus State of the art 🧬 AlphaFold 3: 76% accuracy in protein-small molecule interactions (benchmark test) Vs🎯 State of the art: 52% accuracy 🧬 AlphaFold 3: 65% accuracy in protein-DNA binding prediction Vs 🎯 state of the art: 28% accuracy AlphaFold 3 can’t get researchers all the way to a new drug. All of a new drug’s expected interactions in the body must be proven in wet lab experiments - but this is gonna definitely accelerate the early stage drug discovery phase to accelerate the number of compounds tested in wet lab on high-throuput in vitro models ! Exciting times ! See below an example of problem solving of Alphafold 👇

144

13 Comments

What type of AI would be helpful? The accurate ones.

3

Agents need 3 types of memories to provide a great UX: 1. Chat history i.e. previous messages in that session 2. Personalized user memories 3. Session Summaries With phidata we've finally cracked the holy grail of agentic memory. Video coming soon, but in the meantime, check the code or docs for more info code: https://lnkd.in/dbeGcNT4 docs: https://lnkd.in/d_Dgqbhi

51

Last week Grammarly released a new tool that, according to the company, can detect the overall percentage of AI-generated content in a body of text and even identify which elements were likely created by AI: https://lnkd.in/deZxcZdn While Grammarly appears to primarily target the education segment, I thought it would be interesting to see how people use generative AI for their LinkedIn content. Especially for contributions to the “Collaborative Articles” which can earn a gold “Top Voice” badge identifying an individual as an expert in a particular skill. Full disclosure, I have used ChatGPT, Gemini, and Grammarly to “enhance” my own content. Take my profile summary, Grammarly’s AI detector thinks its about 19% AI-generated, and that’s probably about right. For this experiment, I picked a random topic in the “Leadership” skill. It had a total of 37 contributions at the time. Grammarly tool was able to analyze 27 (some posts were too short or not written in English). Out of those: • 13 contributions were identified as 100% AI-generated • Out of those, seven were authored by users with “Top Voice” badges • Five contributions were identified as having content between 50% and 99% AI-generated. • All were authored by “Top Voices”. Three had the score of 99%. • Nine scored as having 50% or less AI-generated content • Only four were identified as having 0% AI-generated content. None of which were from “Top Voices”. I know that Grammerly’s tool is not perfect, and I am not suggesting that we shouldn’t be using generative AI to enhance our productivity and improve outcomes. I use them daily. I am also willing to concede that as long people find the content useful and helpful, may be it is ok that it is not original? Few other questions come to mind: • What does it mean to be an “expert” in the age of AI? • Do we care how “expertise” is measured and earned in this professional community? • Are we OK with our “Top Voices” coming from ChatGPT Vs. individual's knowledge and experience? • And of course, where are the boundaries of ethical use of AI? What do you think? P.S.: This post is 100% human generated #AI #GenAI #ethics #technology

43

21 Comments

AI memory is fundamental for conversational AI and AI agents. It is critical for building personalized experiences and improving reasoning. Check out Letta’s course to understand how to use AI memory to enhance the customization and quality of your AI app.

15

“But let's not kid ourselves: we need data engineers, SUBJECT MATTER EXPERTS, and people who UNDRSTAND #ONTOLOGIES to build the Knowledge Graphs that will feed the #AI Assistants.” If you need a healthcare data SME who understands ontologies to support your AI endeavors, DM me. I’d love to put two decades of experience in these areas to good use!

8

Unlike direct computer control, this feature is actually available in the Claude web app.

4