Inspiration
I built Mithram(മിത്രം means Friend in Malayalam) because I saw firsthand how clinicians struggle with overwhelming amounts of complex patient data. I wanted to create a tool that transforms raw, intricate information into clear, actionable insights, as reliable as a trusted friend.
What it does
Mithram functions as a Clinical Decision Support Hooks service by integrating directly with healthcare systems and processing patient data through a highly technical, multi-stage analysis pipeline. At its core, Mithram leverages a revolutionary new protocol for Medical AI analysis named GENESIS protocol - Generative Enrichment via NFT and Synthesis, which combines dynamic prompt generation and multi-perspective AI analysis. I created GENESIS protocol during this hackathon.
Mithram dynamically analyzes the patient history by generating four distinct analytical prompts using a zero-temperature GPT model for deterministic outputs. These prompts are executed concurrently via parallel processing, enabling the system to perform in-depth assessments across multiple medical perspectives. The outputs are then synthesized into a unified JSON structure that details the patient overview, recommendations such as patient engagement and interdisciplinary coordination, and identified risk factors.
In Mithram, every patient is tokenized as a unique NFT(contract deployed on Base Sepolia network) following the ERC721 standard. This means that each patient is assigned a digital identity, a patient NFT, that encapsulates their clinical record in an immutable and verifiable way. Every time Mithram performs an analysis on a patient’s data through the GENESIS protocol, the result is not only generated in real time but also stored as metadata. This metadata is minted as part of a broader NFT-based record and is securely kept on decentralized storage (IPFS via Filebase).
Each analysis, stored as metadata on IPFS, links directly back to the associated patient NFT, creating a tamper-proof chain of clinical insights. Anytime a new analysis is done on the same patient, the metadata is updated and new AI analysis on IPFS is linked backed to the previous metadata. This design ensures that every evaluation is both traceable and verifiable, providing an auditable history of the patient's medical data. By combining ERC721 patient NFTs with analysis metadata stored on IPFS, Mithram leverages blockchain principles to ensure data integrity, security, and immutability. This integration offers clinicians a robust, secure framework to track and review patient analyses over time, all while preserving privacy and compliance with healthcare standards.
No PII is stored on-chain. Anonymized data is used to perform AI analysis and linked to patient NFT as metadata(stored on IPFS via Filebase). These metadata have restricted access and can be accessed only via SMART on FHIR Authentication.
How we built it
Backend:
- TypeScript
- Express.js
- LangChain
- FHIR Kit Client
- LavinMQ (Handles asynchronous analysis queue)
- Base Sepolia Network
Frontend:
- React
- Wouter (Routing)
- Tailwind CSS
- Heroicons
- Dexie
Challenges we ran into
One of the toughest hurdles was mastering the FHIR standard to ensure comprehensive and compliant integration of clinical data. I dedicated time to learning FHIR fundamentals from Principles of Health Interoperability by Tim Benson and Grahame Grieve, and deepened my understanding by continuously referring to the official FHIR documentation. This foundational knowledge was critical for implementing secure CDS Hooks services and aligning with modern healthcare interoperability requirements.
Accomplishments that we're proud of
One of my proudest achievements is the development of the GENESIS protocol, a novel approach to AI-driven medical data analysis inspired by the GenRead research (https://arxiv.org/abs/2209.10063). With GENESIS, I fused blockchain technology, IPFS, multi-perspective AI agents, and dynamic prompt generation into a cohesive system that redefines clinical decision support.
In Mithram, every patient is tokenized as a unique ERC721 NFT(https://sepolia.basescan.org/address/0xa4a2708c3dbdeba5d5c2eb576b507427c4a4396c), establishing an immutable digital identity without storing any PII on-chain. Anonymized patient data is used for AI analysis, and the resulting insights are stored as metadata on IPFS via Filebase. This metadata, linked directly to the corresponding patient NFT, forms a secure, tamper-proof chain of clinical insights that can only be accessed through SMART on FHIR authentication.
This integration ensures that each analysis is dynamic, multi-faceted, and securely traceable, a significant leap forward in how medical data is processed and utilized. I'm excited about the potential of GENESIS to enhance the reliability, security, and innovation of clinical decision support systems.
As part of the hackathon, I have also published a research paper on GENESIS Protocol:
Mithram's smart contract is deployed and verified on Base Sepolia TestNet.
Mithram along with GENESIS protocol also opens doors to a secure world of decentralized health analysis using AI and making sure the analysis done using AI is auditable on a public blockchain network. It further opens doors to a novel decentralized way of open research on top of such AI analysis checkpointed on blockchain that seamlessly extends the FHIR to the world of decentralized networks. Privacy of patient data is critical when it comes to AI and blockchain and I believe Mithram takes us closer to a world where we can do that on top of public blockchains.
Mithram does not store any PII Patient data on any database nor does it perform any conventional RAG on the data it deals with. Only non-PII AI analyses are checkpointed on public blockchain with privacy using private IPFS buckets that can be accessed only via SMART on FHIR on Mithram.
What I learned
This project deepened my understanding of integrating disparate systems like blockchain, AI and FHIR-based data handling. I also learned a lot about the FHIR standard as this is the first time I am working on it.
What's next for Mithram - AI Powered Clinical Decision Support System
To evolve Mithram further into a on-chain agentic network of AI doctors that work 24x7 to perform analysis on anonymized data from various similar data points to create and reveal insights like no other system could.
Log in or sign up for Devpost to join the conversation.