Baton

AI-powered institutional memory for volunteer-run organizations—answer questions from Slack history with receipt-backed citations and generate structured handover packs.

Comment

Baton

Inspiration

Volunteer-run organizations depend on people, not documentation. When committee members graduate or step down, valuable knowledge is often buried in Slack conversations or lost entirely. New volunteers spend hours searching for past decisions, contacts, and event details.

We built Baton to solve this problem by turning Slack into an AI-powered institutional memory that provides transparent, receipt-backed answers and generates structured handover documents.

What it does

Baton is an AI-powered Slack Assistant that helps organizations:

  • Answer questions using Slack workspace history
  • Cite the original Slack messages for every answer
  • Generate structured handover packs for outgoing volunteers
  • Preserve institutional knowledge across committee transitions

Instead of guessing, Baton searches Slack, reasons over the results, and returns verifiable answers with links to the original conversations.

How we built it

The application is built with Python using Slack Bolt in Socket Mode.

When a user asks a question:

  1. Slack Assistant receives the request.
  2. Baton searches the workspace using the Slack Real-Time Search API.
  3. Relevant conversations are passed to Groq Llama 3.3 70B.
  4. The model generates a concise answer with citations.
  5. Baton can also generate structured handover documents summarizing responsibilities, contacts, decisions, and useful resources.

The project is deployed on Railway with a lightweight health endpoint for production monitoring.

Challenges we ran into

The biggest challenge was integrating Slack's Assistant workflow with real-time search while keeping responses transparent and trustworthy.

Other challenges included:

  • Managing Slack Socket Mode in production
  • Designing reliable citation formatting
  • Handling missing or incomplete search results gracefully
  • Creating structured handover documents from unstructured conversations

What we learned

Building Baton taught us how to design AI systems that prioritize trust over confidence. Rather than generating unsupported answers, every response is backed by searchable evidence from Slack.

We also gained hands-on experience with Slack's Assistant platform, production deployment, and integrating LLM reasoning into real-world workflows.

What's next

Future improvements include:

  • Multi-workspace support
  • Smarter long-term memory
  • Team analytics and knowledge insights
  • Richer Slack Canvas handover documents
  • Support for additional collaboration platforms

Built With

Share this project:

Updates