• 

Company Brain

AI-powered enterprise memory — documents, knowledge graphs, timelines, and semantic search in one unified platform.

---

Overview

Company Brain is a centralized shared-memory system for the enterprise. It ingests documents, extracts entities and relationships into a knowledge graph, logs chronological events, and lets you query everything through an intelligent conversational interface — powered by NVIDIA NIM for LLM inference and Graphiti for graph-based knowledge extraction.

---

Features

Feature	Description
** Document Management**	Upload PDF, Markdown, or text files. Automatic OCR and text extraction.
** Knowledge Graph**	Interactive force-directed canvas graph. Pan, zoom, drag nodes. Click for details.
** AI Chat**	RAG-powered Q&A. Ask questions, get answers synthesized from documents + graph + timeline.
** Timeline**	Chronological event log with auto-extracted date references from uploaded documents.
** Semantic Search**	Vector embeddings via ONNX or hash-based fallback. Retrieval-augmented generation.
** Role-Based Access**	JWT authentication with Admin, Manager, and Employee roles.
** Entity Extraction**	Automatic capitalized-phrase extraction from document text into graph nodes.

---

Tech Stack

Layer	Technology
Frontend	Next.js 16 (Turbopack), React 19, TypeScript, Tailwind CSS 4, Canvas API
Backend	Python 3.12+, FastAPI, SQLAlchemy, Pydantic
LLM Provider	NVIDIA NIM (meta/llama-3.1-8b-instruct)
Vector Store	ChromaDB (ONNX embeddings with hash fallback)
Graph Database	Neo4j + Graphiti (with local in-memory fallback)
Relational DB	PostgreSQL 16 (with SQLite fallback)
Auth	JWT (python-jose, bcrypt)
PDF Parsing	pypdf

---

Project Structure

companybrain/
├── backend/
│   ├── app/
│   │   ├── api/                    # REST endpoints
│   │   │   ├── auth.py             # JWT login/register
│   │   │   ├── documents.py        # Upload, list, delete
│   │   │   ├── graph.py            # Knowledge graph data
│   │   │   ├── search.py           # Hybrid search
│   │   │   └── timeline.py         # Timeline CRUD
│   │   ├── core/
│   │   │   ├── config.py           # Settings & env vars
│   │   │   ├── database.py         # SQLAlchemy engine
│   │   │   └── security.py         # JWT creation/validation
│   │   ├── models/models.py        # SQLAlchemy models
│   │   ├── schemas/schemas.py      # Pydantic schemas
│   │   └── services/
│   │       ├── chroma_service.py   # Vector embeddings & search
│   │       ├── graphiti_service.py # Graph extraction & storage
│   │       ├── nim_service.py      # NVIDIA NIM LLM client
│   │       └── search_service.py   # Hybrid retrieval pipeline
│   └── run.py                      # Uvicorn entrypoint
├── frontend/
│   └── src/
│       ├── app/
│       │   ├── chat/               # AI Chat interface
│       │   ├── dashboard/          # Stats & quick search
│       │   ├── documents/          # File upload & management
│       │   ├── graph/              # Canvas knowledge graph
│       │   ├── timeline/           # Event timeline
│       │   └── login/              # Authentication
│       ├── components/             # Sidebar, Header, Layout
│       └── utils/api.ts            # API client & types
└── .env                            # Configuration

---

Getting Started

Prerequisites

Dependency	Version	Required
Python	3.12+	Yes
Node.js	20+	Yes
PostgreSQL	16+	Yes
Neo4j	5.x	Optional (local graph fallback available)
NVIDIA API Key	—	Optional (local summary fallback available)

Installation

# 1. Clone
git clone <repo-url>
cd companybrain

# 2. Backend setup
cd backend
python -m venv venv
venv\Scripts\activate    # Windows
# source venv/bin/activate  # macOS/Linux
pip install -r requirements.txt

# 3. Frontend setup
cd ../frontend
npm install

# 4. Configure environment
# Edit .env — DATABASE_URL, NEO4J_URI, NVIDIA_API_KEY

Running

Start both servers in separate terminals:

# Terminal 1 — Backend
cd backend
python run.py
# → http://localhost:8000

# Terminal 2 — Frontend
cd frontend
npm run dev
# → http://localhost:3000

Note: The frontend dev command auto-sets NODE_OPTIONS=--max-old-space-size=8192 to prevent heap OOM during Turbopack compilation.

Default Accounts

Username	Password	Role
admin	admin123	Admin
manager	manager123	Manager
employee	employee123	Employee

---

API Overview

Method	Endpoint	Auth	Description
POST	/api/auth/register	—	Register new user
POST	/api/auth/login	—	Login, get JWT
GET	/api/auth/me	JWT	Current user info
POST	/api/documents/upload	Admin/Manager	Upload file
GET	/api/documents/	JWT	List documents
DELETE	/api/documents/{id}	Admin/Manager	Delete document
POST	/api/search	JWT	Hybrid search + AI answer
GET	/api/graph	JWT	Graph visualization data
GET	/api/timeline/	JWT	Timeline events
POST	/api/timeline/events	JWT	Add timeline event

---

Key Design Decisions

• NVIDIA NIM Only — The sole LLM provider. No Ollama fallback. Chat completions use integrate.api.nvidia.com/v1; embeddings are handled locally via ONNX or deterministic hash functions.
• Graceful Degradation — Every data store has a fallback: Neo4j → local in-memory graph, NVIDIA → local summary, PostgreSQL → SQLite, ONNX → hash embeddings.
• Graph Entity Extraction — When Graphiti/Neo4j is unavailable, the system uses regex-based capitalized-phrase extraction seeded by a shared _graph_node_counter for stable IDs across restarts.
• Canvas Rendering — The knowledge graph viewer uses a continuous requestAnimationFrame render loop with mutable refs and a separate force-simulation loop for smooth interactivity without React re-render overhead.
• Startup Resilience — Previously indexed documents are rebuilt into the local graph automatically on server startup via seed_database().

---

Environment Variables (.env)

# JWT
SECRET_KEY=your_secret_key
ACCESS_TOKEN_EXPIRE_MINUTES=1440

# PostgreSQL
DATABASE_URL=postgresql://user:pass@localhost:5432/companybrain

# Neo4j
NEO4J_URI=neo4j://127.0.0.1:7687
NEO4J_USER=neo4j
NEO4J_PASSWORD=your_password

# NVIDIA NIM
NVIDIA_API_KEY=nvapi-...
NVIDIA_NIM_MODEL=meta/llama-3.1-8b-instruct

# ChromaDB
CHROMA_PERSIST_DIRECTORY=./data/chroma

Built With

• css
• javascript
• python
• typescript