-
Notifications
You must be signed in to change notification settings - Fork 10
Expand file tree
/
Copy pathllms.txt
More file actions
49 lines (28 loc) · 2.37 KB
/
llms.txt
File metadata and controls
49 lines (28 loc) · 2.37 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
Precision Medicine MCP (PatientOne)
An AI-orchestrated framework for multi-modal bioinformatics analysis focusing on a Stage IV Ovarian Cancer case study. This repository demonstrates the use of the Model Context Protocol (MCP) to "stitch" together clinical, genomic, and spatial data.
Core Project Goals
PatientOne Case Study: A synthetic clinical scenario used to teach AI-orchestrated precision medicine.
Modular MCP Architecture: 15 specialized servers acting as domain experts for specific data modalities.
Human-in-the-Loop (CitL): Explicit validation steps for oncologists and clinicians within the AI workflow.
Teaching/Example First: Optimized for clear documentation, persona-based navigation, and low-cost "DRY_RUN" execution.
Key Repository Structure
/servers/: The logic for the 15 custom MCP servers.
/docs/: Human-consumable documentation.
patient-one-tutorial.md: Narrative walkthrough of the cancer case study.
architecture.md: Visual map of the MCP orchestration and SSE transport on GCP.
/data/: Anonymized synthetic datasets for "PatientOne."
/docs/prompts/: Versioned system prompts for medical reasoning and CitL validation.
/infrastructure/: Deployment scripts for GCP Cloud Run and Docker-based local development.
Technical Stack
Protocol: Model Context Protocol (MCP) using SSE (Server-Sent Events) transport.
Cloud: Google Cloud Platform (GCP) - Cloud Run, Cloud Functions.
AI Orchestration: Gemini 3.0 / Claude 4.5 Opus acting as the central "Stitcher."
Bioinformatics: STAR Aligner, TCGA analysis, Visium spatial transcriptomics processing.
Prompting Guidelines for AI
Domain: When generating code or analysis, adhere to HIPAA-principled data handling (even for synthetic data).
Style: Maintain a "Co-pilot" persona—prioritize Clinician-in-the-Loop validation over fully autonomous decision-making.
Orchestration: Always consider how a request might be distributed across the 15 existing MCP servers before proposing new standalone logic.
Why this is effective
Context Efficiency: Instead of an LLM reading every README.md and script, it reads this single file to build a mental model.
Navigation: It explicitly links the /servers and /prompts folders, which are the most important for an AI developer tool to understand.
Persona Alignment: It reinforces the "Co-pilot" and "Teaching" nature of the repo, preventing the AI from suggesting overly complex or "black box" solutions.