HukumPekerja
An Indonesian employment-law information chatbot that uses retrieval-augmented generation to answer questions with supporting references. The implementation uses pgvector search, BGE-M3 embeddings, and selectable Llama 3 / Gemini models. Describe contributions accurately if part of a team.
Team project. The repository lives under the Bideng-Warrior organization rather than my personal account.
Problem
Indonesian workers with labor-law questions face dense statutes and pay-per-consult lawyers; plain-language answers with the actual legal basis are hard to get.
Approach
A RAG pipeline: employment-law documents are embedded with BGE-M3 (via Hugging Face) and stored with pgvector on PostgreSQL; at query time the relevant passages are retrieved and handed to a selectable Llama 3 or Gemini model, so answers cite the law they rest on instead of free-generating. Next.js with Drizzle ORM serves the chat interface and session persistence.
Implemented features
- Retrieval-augmented answers with supporting legal references
- pgvector semantic search over BGE-M3 embeddings
- Selectable Llama 3 / Gemini generation backends
- Conversation history persisted per session
Availability
Repository and public demo are live.