A citation-grounded RAG assistant built over 467 USCIS documents, with speech input and output.
01Overview
ImmigrationBot is a retrieval-augmented conversational assistant built over 467 USCIS documents. It retrieves relevant source material, generates grounded responses and presents citations so users can verify information against official documents. The system also supports speech input and output, making complex immigration information easier to explore through a conversational interface.
02What it does
- 1Embedding-based retrieval over 467 USCIS documents
- 2Grounded generation with visible citations
- 3Speech input and spoken responses
- 4Verifiable answers traceable to official sources
03The challenge
Ensuring every generated answer stayed traceable to official source passages — prioritizing citation fidelity over fluent-but-ungrounded responses.
04The outcome
A citation-first assistant that makes dense government documentation explorable through conversation.
05Built with
- Python
- FastAPI
- LangChain
- ChromaDB
- Groq
- Llama 3.3
- Embeddings
- Speech-to-Text
- Text-to-Speech
Next project — N°07
ClassTrack
A class monitoring and makeup management system for the CSE department at Daffodil International University — tracking whether routine classes are actually held, and handling the follow-up when they aren't.
