How karimulislambd/agentic-research-assistant works
This is a Streamlit research assistant for querying uploaded papers. Users upload PDFs, which are parsed into citable chunks and indexed; questions enter a Groq-backed agent that can search the papers or the web, then the UI displays its answer and citations. A separate LLM judge scores faithfulness and relevance. The graph follows the shown calls and README-described capabilities; unsampled details are not expanded beyond documented behavior.
Main entry points: app.py Streamlit chat · core.py Tool-calling loop · config.py Runtime configuration · ingest.py PDF ingestion
Components 14
User interface
- Streamlit chat
app.py
Paper retrieval
- PDF ingestion
rag/ingest.py - PDF parsing and chunking
rag/ingest.py - Embedding and FAISS store
rag/store.py
Agent and tools
- Tool-calling loop
agent/core.py - System prompts
agent/prompts.py - Tool dispatcher
agent/tools.py - Paper search
agent/tools.py - Web search
agent/tools.py
Answer evaluation
- LLM answer judge
evaluation/judge.py
Other components
- Researcher
- Runtime configuration
config.py - Groq API
- Public web
Connections 20
- Researcher → Streamlit chat: asks and uploads
README.md - Streamlit chat → PDF ingestion: ingests PDFs
app.py - PDF ingestion → PDF parsing and chunking: parses and chunks
rag/ingest.py - PDF ingestion → Embedding and FAISS store: adds chunks
rag/ingest.py - Streamlit chat → Tool-calling loop: submits question
app.py - Tool-calling loop → System prompts: uses prompt
agent/core.py - Tool-calling loop → Tool dispatcher: dispatches tools
agent/core.py - Tool dispatcher → Paper search: routes search
agent/tools.py - Tool dispatcher → Web search: routes search
agent/tools.py - Paper search → Embedding and FAISS store: searches passages
agent/tools.py - Web search → Public web: requests results
agent/tools.py - Tool-calling loop → Groq API: requests completions
agent/core.py - Streamlit chat → LLM answer judge: evaluates answer
app.py - LLM answer judge → System prompts: uses judge prompt
evaluation/judge.py - LLM answer judge → Groq API: requests evaluation
evaluation/judge.py - Tool-calling loop → Runtime configuration: reads settings
agent/core.py - Streamlit chat → Runtime configuration: checks settings
app.py - PDF ingestion → Runtime configuration: reads chunk settings
rag/ingest.py - Embedding and FAISS store → Runtime configuration: reads model settings
rag/store.py - LLM answer judge → Runtime configuration: reads judge settings
evaluation/judge.py