[Contributor] RAG-powered Document Q&A Agent using uAgents + Langchain + Gemini

[ashishsoni-ai]

Task description

Agent Proposal

Proposed folder: contributors/rag-document-qa-agent/
Category: RAG, LLM, Integration
Difficulty: Intermediate

---

What this agent does

A Retrieval-Augmented Generation (RAG) agent that:

• Accepts a PDF or plain-text document as input
• Chunks and embeds it using HuggingFace sentence-transformers
• Stores vectors in ChromaDB
• Answers natural language questions about the document via uAgents chat protocol
• Uses Gemini 2.0 Flash (free tier) as the LLM backbone

---

Why it's useful

Most existing examples focus on live API queries or web scraping.
There is no contributor example showing how to do document-grounded Q&A inside the uAgents ecosystem.
This fills a practical gap — useful for students, researchers, and anyone who wants to chat with their own documents via an agent.

---

Tech Stack

• uagents — agent runtime + chat protocol
• langchain + langchain-google-genai — RAG chain
• chromadb — vector store
• sentence-transformers (HuggingFace) — embeddings
• pypdf — PDF loading
• Gemini 2.0 Flash — LLM (free tier, no cost barrier for contributors)

---

Proposed Folder Layout

contributors/rag-document-qa-agent/
README.md
requirements.txt
.env.example # GEMINI_API_KEY, AGENTVERSE_API_KEY
agent.py # uAgent with RAG chain
ingest.py # Document ingestion + embedding
assets/
demo.png

---

Acceptance Criteria

• Agent registers on Agentverse and responds via chat protocol
• Accepts document path via .env or startup config
• Answers questions grounded in document content (not hallucinated)
• README covers setup, env vars, run steps, and expected output
• .env.example included, no real keys committed
• ruff check . and ruff format . pass
• contributors/CHANGELOG.md updated
• Demo screenshot included

---

About Me

• B.Tech AI & ML student (GGSIPU, 2028)
• Built PDF Chat RAG system: github.com/ashishsoni-ai/pdfchat-groq-rag
• Stack experience: LangChain, ChromaDB, FastAPI, Gemini, HuggingFace, uAgents (learning)
• Contributing via GSSoC 2025

I would like to be assigned this issue and will open a PR within 4 days of assignment.

Target folder (if applicable)

contributors/rag-document-qa-agent/

Contributor checklist

• I will place new agents under contributors/<agent-name>/ only
• I will update contributors/CHANGELOG.md for non-doc changes
• I have starred this repository

[PiyushBorban49]

Hi @fetchai, I am a GSSoC 2026 contributor and would like to work on this.

[princeyadaviiit]

Hi @fetchai, I’d love to work on this RAG Document QA Agent contribution.

This is a very practical addition to the repository because it demonstrates a complete document-grounded conversational workflow inside the uAgents ecosyste something currently missing from the existing examples.

I can implement:

• PDF/text document ingestion pipeline using pypdf
• Chunking + embeddings with sentence-transformers
• ChromaDB vector storage integration
• RAG-based QA flow using LangChain
• Gemini 2.0 Flash integration as the LLM backend
• uAgents chat protocol support for interactive querying
• Configurable document loading via .env or startup configuration
• Proper project structure under:
  contributors/rag-document-qa-agent/

I’ll also ensure:

• Clean documentation with setup and usage instructions
• .env.example included with no exposed secrets
• ruff check . and ruff format . compliance
• contributors/CHANGELOG.md updated
• Demo screenshot added

I already have hands-on experience with LangChain, ChromaDB, Gemini, HuggingFace embeddings, and PDF-based RAG systems, so I should be able to complete this efficiently.

I’m contributing via GSSoC and can open a PR within 4 days of assignment.

Please assign this issue to me. Thanks!

[AKIB2005]

Hi @ashishsoni-ai / @fetchai team! 👋
I'm a GSSoC '26 contributor and I'd love to work on this issue!
My planned implementation:
ingest.py — Document ingestion pipeline:

Load PDF/text using pypdf or plain file read
Chunk document using LangChain's RecursiveCharacterTextSplitter
Generate embeddings using sentence-transformers (HuggingFace all-MiniLM-L6-v2)
Store vectors in ChromaDB with persistent local storage

agent.py — uAgent with RAG chain:

Register agent on Agentverse using uagents chat protocol
On receiving a question message, retrieve top-k relevant chunks from ChromaDB
Pass retrieved context + question to Gemini 2.0 Flash via langchain-google-genai
Return grounded answer via chat protocol response

README.md — Full setup guide covering env vars, run steps, and expected output with demo screenshot
Acceptance Criteria I'll fulfill:

good first issue — well-scoped Python agent, no complex infra needed
help wanted — fills the document Q&A gap in the uAgents ecosystem
Agent registers on Agentverse and responds via chat protocol
.env.example included, no real keys committed
ruff check . and ruff format . pass
contributors/CHANGELOG.md updated
Demo screenshot included

Could you please assign this to me? I'll submit the PR within 4 days! 🚀

[apoorvakappu123-blip]

Hi @ashishsoni-ai

I would like to work on this issue under GSSoC'26.

I’ve reviewed the proposed architecture and acceptance criteria, and I’m interested in implementing the RAG-powered Document Q&A Agent using uAgents, LangChain, ChromaDB, HuggingFace embeddings, and Gemini 2.0 Flash.

My implementation plan includes:

• Document ingestion pipeline for PDF/text files using pypdf
• Chunking and embedding with sentence-transformers
• Persistent vector storage using ChromaDB
• Retrieval-based QA workflow with LangChain
• Gemini 2.0 Flash integration for grounded responses
• uAgents chat protocol support for interactive querying
• Proper setup documentation, .env.example, demo screenshot, and changelog updates

I’ll ensure the implementation follows the repository structure and passes ruff check . and ruff format ..

Looking forward to contributing

[TAKI-074]

Hi, I’m a GSSoC 2026 contributor. I would like to work on this issue. Can you please assign it to me?

[sathyashreekv]

i would like to work on this issue please assign to me

[indhumitha]

Hi @ashishsoni-ai
I am a contributor and I would like to contribute to this issue as part of GSSoC'2026. Please assign.