Inspiration

Incident response and compliance audits require reliable, auditable collections of evidence. Teams waste hours gathering scattered logs, PR diffs, chat threads and transcripts after incidents. EvidenceBot uses LLM-driven automation and authenticated tool access to assemble tamper-evident evidence bundles so teams can move faster and reduce post-incident friction.

What it does

  • Connects to user-scoped tools (GitHub, Slack, PagerDuty, Drive) via Composio.
  • Listens to triggers (commits, alerts, messages) and normalizes events.
  • Collects artifacts (diffs, logs, threads, files) using authenticated tool calls or proxy requests.
  • Assembles a signed evidence bundle (manifest + ZIP + HMAC) and stores it securely.
  • Presents a polished UI for onboarding, incident review, artifact preview, and bundle download.

How we built it

  • Frontend: React + Vite + TypeScript + Tailwind for a clean, startup-grade UI.
  • Backend: Node.js + TypeScript, modular services for Composio integration, webhook handling, worker queue, collector and assembler jobs.
  • Storage & DB (demo): JSON-backed local persistence; recommended production stack is Supabase (free tier) for Postgres + Storage.
  • Local dev: mock Composio implementation to iterate without real credentials; demo seed data and simple worker loop for end-to-end flow.

Challenges we ran into

  • Securely handling per-user OAuth flows while keeping the demo friction low; solved with Composio and a mock mode.
  • Normalizing heterogeneous trigger payloads into a canonical event model for consistent downstream processing.
  • Managing large artifacts and file transfers; demo uses inline storage but production needs streaming and retention policies.

Accomplishments that we're proud of

  • End-to-end demoable flow: onboarding, webhook ingestion, artifact collection and evidence storage in demo mode.
  • Robust Composio wrapper with a mock implementation enabling rapid development and safe demos.
  • Clean, minimal API and job worker design that can be extended to production services (Supabase/S3, Redis, background workers).

What we learned

  • Composable tool-auth layers (Composio) dramatically reduce integration complexity and security surface for agentic actions.
  • Building strong normalization early prevents downstream edge cases and simplifies templates for artifact collection.
  • Demo-mode mocks are essential to iterate on UI/UX and stakeholder demos before provisioning cloud resources.

What's next for Compliance Evidence Collector

  • Implement assembler: manifest generation, zip creation, HMAC signing, and presigned download endpoints.
  • Replace local JSON persistence with Supabase Postgres + Storage and add RBAC and audit logs.
  • Add additional connectors (Datadog/CloudWatch via proxy, Google Drive transcripts) and improve large file streaming.
  • Harden security: token rotation, least-privilege scopes, audit search and evidence retention policies.
  • Complete frontend demo: ConnectCard, dashboard, event detail with artifact previews and signature verification.

Built With

Share this project:

Updates