Commitment Compiler

Don’t just chat—commit. Our AI compiler turns unstructured conversations into binding contracts with verifiable owners, deadlines, and clear deliverables.

Comment

Commitment Compiler - Devpost Submission

Inspiration

We’ve all been there: a Slack thread ends with "I’ll try to get to it," and a week later, nobody remembers who was supposed to do what. In the move to remote and async work, our primary interface for collaboration—chat—has remained fundamentally unstructured. We were inspired to build a tool that treats human conversation like a protocol. Just as a compiler turns messy source code into a binary contract, we wanted a tool that turns messy human dialogue into a structured Execution Contract.

What it does

Commitment Compiler is a 5-stage "linguistic compiler" that audits conversation threads (Slack, Email, Discord) to identify and secure accountability.

  • Smart Audit Profiles: Optimized templates for Slack threads, Email chains, Sprint Planning, and Bug Fixes to handle the unique linguistic nuances of different mediums.
  • Stage A: Speech Act Linter: Classifies every message using pragmatic theory and performs token-level verification, highlighting "hedges" directly in the source thread.
  • Stage B: 4-Pillar Extraction: Isolates the Owner, Action, Deadline, and Deliverable for every commitment.
  • Stage C: Verifier Pass: Cross-references every commitment with the raw thread to find counter-evidence or "weak" support.
  • Stage D: Risk Analyzer: Surfaces "Silent Killers" like unacknowledged asks, role mismatches, and vague deadlines (the dreaded "ASAP").
  • Stage E: Dynamic Clarifier: Generates natural, context-aware AI questions to resolve ambiguity before it becomes a blocker.

The result is a deterministic Agreement Health Score and a downloadable Interaction Receipt.

How we built it

  • Backend: FastAPI + Python powered by a multi-stage GPT-4o pipeline. We used Pydantic 2.0 for strict schema enforcement at every stage of the "compilation."
  • Frontend: A high-fidelity Next.js 14 dashboard using Tailwind CSS and shadcn/ui.
  • Animations: Framer Motion for the interactive "Agreement Health" gauge and evidence highlighters.
  • DevOps: uv for lightning-fast Python dependency management and reliable builds. Configured for production with specialized Procfile and Vercel integration.

Challenges we ran into

The biggest challenge was moving beyond simple "keyword matching." We had to teach the engine the difference between a Role (e.g., "I'll lead the report") and a Deliverable (e.g., "I'll send the Q4 Slide Deck"). We solved this by implementing a dedicated Verification Stage (Stage C) and a Dynamic Clarifier (Stage E) that proactively asks for missing deliverables.

Another hurdle was the Health Formula. Initial versions were too harsh. We refactored the math to give partial credit for "Acknowledgments" and "Implied Agreements," creating a more realistic metric that users can actually improve by answering the generated prompts.

Accomplishments that we're proud of

  • Context-Aware Prompts: Our Dynamic Clarifier doesn't just say "Deadline missing," it generates a natural-sounding Slack message you can copy and paste immediately with a "Copy all" feature.
  • Evidence Linking: The UI allows you to click any commitment and see the exact message source highlighted in the thread—zero trust, full verification.
  • Audit Persistence: We implemented a persistent History system and multi-format exports (Markdown/JSON), turning ephemeral chats into durable records.

What we learned

We delved deep into Speech Act Theory (Austin & Searle) to understand how language actually creates obligations. We learned that "accountability" isn't a vibe—it's a data structure. Building this taught us how to chain multiple LLM calls while maintaining state and data integrity across a complex pipeline.

What's next for Commitment Compiler

  • Slack/Discord Bot: Moving from "Paste your thread" to "Real-time audit" where the bot flags vague commitments in the channel as they happen.
  • Calendar Integration: Automatically turning verified deadlines into scheduled tasks.
  • Team Analytics: A "Commitment Velocity" leaderboard to see which teams have the highest "Agreement Health" over time.
  • Multi-Thread Resolution: Connecting commitments made in Slack to the final deliverables uploaded in GitHub or Jira.

Built With

Share this project:

Updates