Back to Projects
RAG System
The Archive AI
The Archive AI is a retrieval-augmented generation platform built to connect user questions with structured data sources before LLM response generation.
Website: https://the-archive-ai-eosin.vercel.app/
GitHub: https://github.com/SawSimonLinn/the-archive-ai
AI Architecture
RAG Pipeline Development
Data Retrieval Design
Accuracy Optimization
Timeline
MVP build
Challenge
Base LLM responses were often too generic or inaccurate when operating without external context and grounded retrieval.
Solution
We implemented a RAG pipeline with embeddings and vector search so answers are generated from relevant retrieved data instead of ungrounded model output.
Tech Stack
- Next.js
- TypeScript
- Embeddings
- Vector Search
- RAG
Outcomes
- Retrieved relevant context before answer generation
- Improved response accuracy by grounding outputs in real data
- Reduced generic and hallucinated responses
- Created a reusable retrieval foundation for future AI features