SciNets

SciNets is a graph-constrained scientific discovery system for literature synthesis, multi-hop reasoning, and hypothesis generation.

Instead of treating scientific exploration as plain retrieval or free-form text generation, SciNets builds a local concept graph from literature, reasons over that structure, and uses the graph to generate more grounded mechanistic hypotheses.

Why SciNets

Scientific questions often require connecting ideas spread across multiple papers rather than summarizing a single source. SciNets is built for that setting.

It focuses on:

• literature-first discovery
• graph-backed multi-hop reasoning
• hypothesis generation with structural grounding
• evidence-aware synthesis rather than unconstrained brainstorming

Core Workflow

User Query
   ->
Literature Retrieval
   ->
Concept / Relation Extraction
   ->
Session-Level Concept Graph
   ->
Multi-Hop Path Reasoning
   ->
Hypothesis + Evidence Synthesis

What the System Does

• Searches and structures scientific literature
• Builds session-level concept graphs from retrieved evidence
• Identifies bridge paths between concepts that rarely co-occur directly
• Generates hypotheses from graph structure and supporting context
• Streams intermediate reasoning activity to the frontend
• Supports authenticated user workflows for interactive discovery

Research Paper

SciNets is based on the paper:

SciNets: Graph-Constrained Multi-Hop Reasoning for Scientific Literature Synthesis
Sauhard Dubey, 2025
Read on arXiv

If you are exploring the repo for research context, this paper is the best high-level description of the motivation, evaluation framing, and system behavior.

Repository Structure

backend/                 FastAPI backend, agent pipeline, tools, auth, storage
scinets-discovery-hub/   React + Vite frontend
docs/                    Architecture and evaluation notes
Dockerfile               Backend deployment image
docker-compose.yml       Local multi-service development setup

Tech Stack

• Backend: Python, FastAPI, LangGraph, Pydantic
• Frontend: React, TypeScript, Vite, Tailwind CSS, shadcn/ui
• Data / infra: PostgreSQL, Docker
• Deployment: Render for backend, Vercel for frontend

Local Development

1. Backend

cd backend
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
python server.py

The backend expects configuration through environment variables. See backend/.env.example.

2. Frontend

cd scinets-discovery-hub
npm install
cp .env.example .env
npm run dev

The frontend expects configuration through environment variables. See scinets-discovery-hub/.env.example.

Environment Variables

Example env files are included:

• backend/.env.example
• scinets-discovery-hub/.env.example

Production secrets should only be set in deployment platform settings and should never be committed to git.

Important production variables typically include:

• backend API keys and auth secrets
• database connection settings
• Google OAuth client configuration
• frontend API base URL

Deployment

SciNets is structured as a split deployment:

• Backend: Docker-based deployment suitable for Render
• Frontend: static Vite deployment suitable for Vercel

The public repository intentionally keeps deployment configuration but excludes private runtime secrets and generated local artifacts.

Documentation

• Architecture Notes
• Evaluation Notes

Current Scope

SciNets is best understood as a research-grade discovery and ideation system, not a fully autonomous scientific truth engine.

It is strongest when:

• the retrieved corpus is compact and relevant
• the reasoning path is structurally meaningful
• a human remains in the loop to validate important conclusions

Repository Status

This public repository is a cleaned version of the project codebase. Sensitive local outputs, generated debug artifacts, and non-essential private runtime files were removed before publication.