Projects
FileLLM - AI-Powered Document & Code Assistant
Full-stack RAG web app bridging document understanding and code editing in a unified interface.
Overview
A full-stack RAG (Retrieval Augmented Generation) web application that bridges the gap between document understanding and code editing within a unified interface. It allows users to upload various document formats and index entire codebases for semantic interaction.
The Problem
Users struggle to get context-aware answers from their documents and need AI assistance to edit codebases without leaving the document context. Existing tools don't bridge the gap between document understanding and code editing.
Approach
- Built a retrieval-augmented generation pipeline to bridge document understanding and code editing.
- Handles PDF, Google Docs (via API), DOCX, and plain text files with automated chunking.
- AI engine uses uploaded documents as live context for generating human-like, accurate responses.
- Maps entire codebases into vector space to enable AI-powered code edits with line-number awareness.
- Leverages ChromaDB for high-speed semantic retrieval across both documents and source code.
Technical Stack
| Layer | Technology |
|---|---|
| Backend | Django 4.2+, Python |
| Vector DB | ChromaDB (Vector Database) |
| AI/ML | Sentence-Transformers, Groq API (Llama models) |
| Integrations | Google Docs API, pypdf, python-docx |
| Frontend | HTML5, CSS3, JavaScript (ES6+) |
What I learned
- Implementing a complete RAG pipeline from scratch (chunking, embedding, retrieval, generation).
- Vector database integration (ChromaDB) for semantic search at scale.
- Django model design for complex many-to-many relationships (projects, documents, chat sessions).
- Building context-aware AI prompts that combine retrieved documents with chat history.
- Semantic code indexing with file tree structure and line-number awareness for precise AI edits.
- Security best practices (environment variables, credentials management, input validation).