LLM powered chatbot for nutrition science information
Project: Nutrition Science Chatbot
Situation
The project needed a way to make a large body of nutrition research easier to access, answer user questions instantly, reduce manual research effort for common topics, and maintain scientific grounding in the responses.
Task
Build an AI-powered chatbot that could ingest nutrition content, retrieve relevant source material, generate useful answers, and keep the experience reliable enough for ongoing user research and discovery.
Action
- Built an automated content pipeline for scraping, cleaning, chunking, and storing nutrition research in a vector database
- Developed a chat interface for natural-language question answering with source citations and conversation history
- Added knowledge management workflows for regular content updates, quality control, and feedback collection
- Integrated OpenRouter for LLM access alongside custom retrieval endpoints, monitoring, and analytics
Result
- Instant access to nutrition research
- 24/7 availability for questions
- Reduced research time for common topics
- Increased content discoverability
Key Innovations:
- Hybrid retrieval-augmented generation
- Context-aware responses
- Cost-effective LLM usage
Technologies
Backend Technologies:
- Python
- Scrapy (web scraping)
- Zyte API (proxy rotation)
- LangChain (AI orchestration)
- ChromaDB (vector database)
- FastAPI (API server)
- Redis (caching)
- Sentry (error tracking)
- Pytest (testing)
Frontend Technologies:
- Next.js
- Tailwind CSS
- React Hook Form
- Vercel (hosting)
AI/ML Components:
- HuggingFace embeddings
- OpenRouter API
- Custom prompt engineering
- Retrieval-augmented generation
APIs & External Services:
- OpenRouter (LLM access)
- Vercel (deployment)
- AWS S3 (document storage)
- GitHub Actions (CI/CD)
Contact us for your project needs.