🧠 Neuro-Symbolic AI Scientist Agent
This project implements a distributed AI Research Scientist designed to accelerate discovery in fields like Alzheimer’s disease (AD) by generating, validating, and designing scientifically rigorous, actionable experiments.
🎯 Project Goal
To automate the full scientific discovery process—from literature retrieval to laboratory blueprint—by combining the creative intuition of Large Language Models (LLMs) with the logical rigor of Symbolic AI.
✨ Key Features & Innovation
| Feature | Description | Sources |
|---|---|---|
| Scientific Rigor (The Symbolic Edge) | Hypotheses are formally validated using the Z3 SMT Solver to check for logical inconsistencies against a knowledge base of known AD facts (e.g., Aβ clearance does not guarantee cognitive improvement). | |
| Actionable Output | The system generates a comprehensive Experiment Blueprint (including required models like 5xFAD mice, detailed treatment groups, specific outcome measures like Morris water maze, and duration). The output is available as JSON, LaTeX, and PDF. | |
| Scalable Architecture | Built on 4 core microservices using Docker and FastAPI, enabling parallel development and independent testing. All communication uses JSON over HTTP. | |
| Advanced LLM Integration | Leverages LLaMA 3 (via the Cerebras API) as the "neural brain" for both creative hypothesis generation (identifying knowledge gaps) and structured experiment synthesis. | |
| Professional Demo | A central Streamlit Dashboard (Port 8501) orchestrates the entire pipeline, visually displaying the step-by-step process from query to final blueprint. |
🏗️ Architecture and Workflow Pipeline
The pipeline is split between two roles: Person A (Neural Intelligence) and Person B (Symbolic & Interface).
| Step | Service Name (Port) | Owner Role | Purpose & Technology |
|---|---|---|---|
| 1. Retrieval | ingest-search (8000) |
Person A | Finds the top 3 relevant Alzheimer’s papers based on a user query using semantic search (FAISS/Chroma). |
| 2. Generation | hypothesis-gen (8001) |
Person A | Identifies a knowledge gap and proposes a testable hypothesis (e.g., "Combining X with Y will improve Z outcomes") using LLaMA 3. |
| 3. Validation | z3-validator (8002) |
Person B | Checks if the hypothesis logically contradicts known biological facts defined in the knowledge base using the Z3 SMT Solver. |
| 4. Design | experiment-design (8003) |
Person B | Converts the validated hypothesis into a detailed experimental blueprint using LLaMA 3. |
| Interface | dashboard (8501) |
Person B | Streamlit UI that calls all services sequentially and displays the final result and download options. |
Log in or sign up for Devpost to join the conversation.