Building KnowledgeSpace AI Agent Together — Security Fixes, Production Updates & Call for Collaborators (GSoC 2026)

Hello @visakh, @Tom, mentors, and everyone in the INCF community,

I hope you’re all doing well! My name is Sandeep Chauhan, and I’ve been spending the past few days diving deep into the KnowledgeSpace AI Agent (Project #20).

I wanted to share some updates with the community — and more importantly, invite others to join and build this together.


What I’ve Been Working On

While testing the agent, I noticed that when errors occur (like rate limits or connection issues), the raw technical details were being shown directly to users — things like LLM model names, billing tiers, Python file paths, and SSL internals.

This isn’t ideal for a production tool that researchers will use daily. So I worked on a fix:

Problem Before After
Rate limit errors 429 + gemini-2.5-flash-lite + quotaValue: 20 “I’m currently experiencing high demand. Please try again in a moment.”
SSL/connection issues _ssl.c:1032 + UNEXPECTED_EOF “Unable to connect to the server. Please check your connection.”
General API errors Raw JSON with internal details Clean, friendly messages

PR #48: fix(frontend): sanitize API error messages to prevent info disclosure by SandeepChauhan00 · Pull Request #48 · INCF/knowledge-space-agent · GitHub
Issue #47: [Security/UX] API Error Messages Expose Internal System Details to Users · Issue #47 · INCF/knowledge-space-agent · GitHub

I also submitted PR #46 to fix the chat history persistence issue (#45) — so conversations now stay saved even after a page refresh.


Where I Think We Can Go Next

I’ve been thinking about what would make this agent truly special for the neuroscience community. Here are some ideas — but I’d love to hear yours too:

  • Streaming responses — so users see answers appear in real-time (like ChatGPT)
  • Confidence scores — showing how certain the agent is about each answer
  • Hybrid search — combining vector, keyword, and graph-based retrieval
  • Offline mode — for researchers with limited internet access
  • Citation support — linking directly to source datasets
  • Dark mode & accessibility — making it usable for everyone

What features would YOU find most useful? I’d love to hear from researchers, students, and contributors alike.


Let’s Build This Together

I’m fully committed to this project for GSoC 2026 — but I don’t want to build alone.

If you’re interested in contributing, learning, or just brainstorming ideas — please drop a comment below! Whether you’re a beginner or experienced, there’s room for everyone.

Some ways you could help:

  • Frontend improvements (React, UI/UX)
  • Backend enhancements (Python, LangGraph, RAG)
  • Testing and finding bugs
  • Documentation and onboarding guides
  • Ideas and feedback

A Humble Request to Mentors

@visakh — I would really appreciate any feedback on the work so far. Also, if the agent is getting close to being ready for a public demo, I’d love to help make that happen.


Thank you all for being part of this amazing community. I’m grateful for the opportunity to contribute, and I truly believe we can make KnowledgeSpace Agent something the entire neuroscience world will benefit from.

Looking forward to hearing your thoughts!

Warm regards,
Sandeep Chauhan
GitHub: SandeepChauhan00


Who’s interested in collaborating? Drop a comment below — even just “I’m in!” works. Let’s build something great together. :raised_hands:

1 Like

Im really excited to build something incredible by collaborating here. If you allow I’ll be in. One more thing since I’m not very experienced it would be great if you make me understand everything clearly. I’m ready to give 100%.

Hey @Deepika_Singh! :wave:

Welcome aboard — really glad to have you here! And please don’t worry about experience level at all. I’m learning too, and that’s exactly what open source is about — we grow by building together.

Here’s how you can get started:

Step 1: Set up the project locally

  • Fork and clone the repo: INCF/knowledge-space-agent
  • Follow the README to get it running on your machine
  • If you run into any setup issues, feel free to ask here or open a GitHub issue — that itself is a valuable contribution!

Step 2: Explore the codebase

  • The frontend is in React, and the backend is Python (FastAPI + LangGraph)
  • Try using the agent, ask it some neuroscience questions, and see how it responds
  • Note anything that feels broken, confusing, or could be improved

Step 3: Pick something to work on
Here are some beginner-friendly areas:

  • Frontend/UI improvements — styling, layout, responsiveness
  • Documentation — improving setup guides, adding screenshots
  • Bug hunting — test the app and report any issues you find
  • Testing — writing test cases for existing features

Check the open issues on GitHub — look for anything that interests you, and drop a comment saying you’d like to work on it.

No pressure, no rush. Even small contributions like fixing a typo in the docs or reporting a bug count. The important thing is to start.

If you get stuck at any point, just ping me here or on GitHub (@SandeepChauhan00) — happy to help!

Let’s build something great .

Quick update on my recent contributions:

Just submitted PR #64 — the first unit test suite for the backend API. It covers all 5 endpoints with 28 tests, all passing. Heavy dependencies like torch and vertexai are mocked so the tests run instantly without any cloud setup.

This builds on top of:

  • PR #48 (error message sanitization)
  • PR #46 (chat history persistence)
  • PR #43 (Python version compatibility)

The project now has a testing foundation that other contributors can build on. If anyone wants to add more tests, the setup in conftest.py makes it straightforward.

Issue: [Testing] Add unit test suite for backend API endpoints and core modules · Issue #63 · INCF/knowledge-space-agent · GitHub
PR: test: add unit test suite for backend API endpoints (#63) by SandeepChauhan00 · Pull Request #64 · INCF/knowledge-space-agent · GitHub

@Deepika_Singh if you’re looking for a good starting point, writing additional tests would be a great way to get familiar with the codebase!