RAGVenture

RAGVenture is an intelligent startup idea generator powered by Retrieval-Augmented Generation (RAG). It helps entrepreneurs generate innovative startup ideas by learning from successful companies, combining the power of large language models with real-world startup data.

Why RAGVenture?

Traditional startup ideation tools either rely on expensive API calls or generate ideas without real-world context. RAGVenture solves this by:

• Completely FREE: Runs entirely on your machine with no API costs - zero API keys required!
• Smart Model Management: Automatically handles model deprecation and failures with intelligent fallback
• Data-Driven: Learns from real startup data to ground suggestions in reality
• Context-Aware: Understands patterns from successful startups
• Intelligent: Uses RAG to combine LLM capabilities with precise information retrieval
• Resilient: Works offline with local models when external APIs are unavailable
• Production-Ready: 177 tests with comprehensive coverage, Docker runtime fixes, and monitoring

System Requirements

• Python 3.11 or higher
• 8GB RAM minimum (16GB recommended)
• 2GB disk space for models and data
• Operating Systems:
  • Linux (recommended)
  • macOS
  • Windows (with WSL for best performance)

Quick Start

1. Installation:

# Clone the repository
git clone https://github.com/valginer0/RAGVenture.git
cd RAGVenture

# Create virtual environment
python -m venv .venv

# Activate virtual environment
# On Windows:
.venv\Scripts\activate
# On Unix or MacOS:
source .venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Install spaCy language model for market analysis
python -m spacy download en_core_web_sm

2. Environment Setup (Optional - system works completely FREE without any setup!):

# Optional: HuggingFace token for enhanced remote models (system works completely FREE without it)
export HUGGINGFACE_TOKEN="your-token-here"  # Get from huggingface.co

# Smart model management (enabled by default)
export RAG_SMART_MODELS=true
export RAG_MODEL_CHECK_INTERVAL=3600
export RAG_MODEL_TIMEOUT=60

# Optional: LangChain tracing (debugging)
export LANGCHAIN_TRACING_V2=true
export LANGCHAIN_ENDPOINT="https://api.smith.langchain.com"
export LANGCHAIN_API_KEY="your-langsmith-api-key"
export LANGCHAIN_PROJECT="your-project-name"

3. Generate Ideas:

# Generate 3 startup ideas in the AI domain
python -m rag_startups.cli generate-all "AI" --num-ideas 3

# Generate ideas without market analysis
python -m rag_startups.cli generate-all "fintech" --num-ideas 2 --no-market

# Check model health and status
python -m rag_startups.cli models status

# Use custom startup data file
python -m rag_startups.cli generate-all "education" --file custom_startups.json

Features & Capabilities

Core Features

• Intelligent Idea Generation:
  • Uses RAG to combine LLM knowledge with real startup data
  • Generates contextually relevant and grounded ideas
  • Provides structured output with problem, solution, and market analysis

Command-Line Interface

Commands:

• generate-all: Generate startup ideas with market analysis
  • Required argument: Topic or domain (e.g., "AI", "fintech")
  • Options:
    • --num-ideas: Number of ideas (1-5, default: 1)
    • --file: Custom startup data file (default: yc_startups.json)
    • --market/--no-market: Include/exclude market analysis
    • --temperature: Model creativity (0.0-1.0)
    • --print-examples: Show relevant examples

Smart Model Management

• Automatic Fallback: Falls back to local models when external APIs fail
• Model Migration Intelligence: Handles model deprecation (e.g., Mistral v0.2→v0.3) automatically
• Health Monitoring: Continuous model health checks and status reporting
• Local Resilience: Works completely offline with local models
• CLI Management: models command for status, testing, and diagnostics

Technical Features

• Smart Analysis:
  • Semantic search for relevant examples
  • Automatic metadata extraction
  • Pattern recognition from successful startups
• Performance Optimized:
  • One-time embedding generation (~22s)
  • Fast idea generation (~0.5s per idea)
  • Efficient data processing (~0.1s load time)
• Production Quality:
  • 31 comprehensive unit tests
  • Automated code formatting
  • Extensive error handling

Performance

Typical processing times on a standard machine:

• Initial Setup: ~22s (one-time embedding generation)
• Data Loading: ~0.1s
• Idea Generation: ~0.5s per idea

Docker Support

For containerized deployment, we provide both CPU and GPU support.

Prerequisites

• Docker and Docker Compose
• For GPU support:
  • NVIDIA GPU with CUDA
  • NVIDIA Container Toolkit
  • nvidia-docker2

Quick Start with Docker

# CPU Version (recommended - fully tested)
docker-compose up app-cpu

# GPU Version (with NVIDIA support)
docker-compose up app-gpu

# Run with custom data file
docker-compose run --rm app-cpu python -m rag_startups.cli generate-all fintech --num-ideas 1 --file /app/yc_startups.json

Docker Status: ✅ Production Ready - All runtime issues resolved, works end-to-end with real data.

Run from GitHub Container Registry (GHCR)

If you prefer pulling a prebuilt image from GHCR:

1. Login to GHCR (needs a GitHub token). Replace USERNAME/TOKEN accordingly.

echo "$GITHUB_TOKEN" | docker login ghcr.io -u <USERNAME> --password-stdin
# or
echo "<TOKEN>" | docker login ghcr.io -u <USERNAME> --password-stdin

2. Pull the image:

docker pull ghcr.io/valginer0/rag_startups:0.9.2
# or latest if available
docker pull ghcr.io/valginer0/rag_startups:latest

3. Run the CLI (using your local .env for tokens/settings):

docker run --rm -it \
  --env-file .env \
  ghcr.io/valginer0/rag_startups:0.9.2 \
  python -m rag_startups.cli generate-all "AI" --num-ideas 2

Tip: For offline/deterministic runs, set HUGGINGFACE_HUB_OFFLINE=1 and TRANSFORMERS_OFFLINE=1 in your .env.

Maintainers – Publish to GHCR:

# After building locally (e.g., docker build -t ghcr.io/valginer0/rag_startups:dev .)
docker tag ghcr.io/valginer0/rag_startups:dev ghcr.io/valginer0/rag_startups:0.9.2
docker push ghcr.io/valginer0/rag_startups:0.9.2
docker tag ghcr.io/valginer0/rag_startups:0.9.2 ghcr.io/valginer0/rag_startups:latest
docker push ghcr.io/valginer0/rag_startups:latest

Development Setup

1. Clone and setup:

git clone https://github.com/valginer0/RAGVenture.git
cd RAGVenture
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

2. Install development dependencies:

pip install -r requirements.txt
pre-commit install  # Sets up automatic code formatting

3. Run tests:

pytest tests/  # Should show 178 passing tests

Testing & Offline Policy

This project enforces fully offline, deterministic tests:

• Tests block outbound HTTP(S) by default via an autouse fixture in tests/conftest.py that patches requests.sessions.Session.request.
• Autouse fixtures also mock model-loading/network paths:
  • huggingface_hub.model_info in rag_startups/cli.py preflight
  • transformers.pipeline at all call sites (e.g., rag_startups.embed_master, rag_startups.core.rag_chain, CLI)
  • huggingface_hub.InferenceClient and the bound imports used by rag_startups/idea_generator/generator.py
  • rag_startups.embed_master.calculate_result is replaced with a deterministic helper during tests
• Offline env vars are forced: HUGGINGFACE_HUB_OFFLINE=1, TRANSFORMERS_OFFLINE=1.
• To explicitly allow network in a specific test, add the marker: @pytest.mark.allow_network.

Runtime (non-test) CLI runs are allowed to use the network and will honor your .env.

Data Requirements

RAGVenture works with startup data in JSON format. Two options:

1. Use YC Data (Recommended):

   • Download from Y Combinator
   • Convert CSV to JSON:

     python -m rag_startups.data.convert_yc_data input.csv -o startups.json

2. Use Custom Data:

   • Prepare JSON file with required fields
   • See docs/data_format.md for schema

Troubleshooting

1. Embedding Generation Time:

   • First run takes ~22s to generate embeddings
   • Subsequent runs use cached embeddings
   • GPU can significantly speed up this process

2. Common Issues:

   • Missing HUGGINGFACE_TOKEN: Sign up at huggingface.co
   • Memory errors: Reduce batch size with --max-lines
   • GPU errors: Ensure CUDA toolkit is properly installed

Documentation

• docs/api.md: API documentation
• docs/examples.md: Usage examples
• docs/data_format.md: Data schema
• CONTRIBUTING.md: Development guidelines

Contributing

See CONTRIBUTING.md for development setup and guidelines.

License

This project is licensed under the MIT License - see LICENSE for details.

Startup Names and Legal Considerations

Name Generation

• Each generated startup name includes a unique identifier (e.g., "TechStartup-x7y9z")
• This identifier ensures technical uniqueness within the tool
• The unique identifier is NOT a substitute for legal name verification

Important Notes for Users

• Generated names are suggestions only
• The uniqueness of a name at generation time does not guarantee its availability
• Users must perform their own due diligence before using any name

Name Verification Resources

• USPTO Trademark Database: https://www.uspto.gov/trademarks
• State Business Registries
• Domain Name Availability Tools
• Professional Legal Counsel

Future Features

• Name availability checking tool (planned)
• Integration with business registry APIs