Relay

Inspiration

Our project solves a real problem that we, SWE interns to PMs face: vague product requirements and incompatible interpretations. Engineers have to decode and figure out:

Clarifying requirements back and forth
Searching which files to change
Writing acceptance criteria
Creating boring boilerplate PRs

What it does

This is how our Agent System works:

PM in Slack: "Fix React version issue"
Postman AI Agent processes the request
Ripgrep finds relevant code files (Settings.tsx, theme.ts)
snowflake cortex ai generates PR code (≤30 lines)
GitHub PR created automatically
Slack notifies team with reasoning trace
Dashboard shows history and the current usage of current pr's and of snowflake in general

Result: PR + acceptance criteria + impacted files in 30 seconds.

Tech Stack

| Core Orchestration -Postman AI Agent + Flows
| Code Search | Ripgrep API (Node.js/Express) | PR Generation | MIstral Large
| PR Creation | GitHub REST API
| Notifications | Slack Webhooks + Block Kit
| Dashboard | Next.js 16 + Express backend
Perfect — here’s your Devpost-ready version (slightly reduced for length, but still polished, professional, and readable for judges). You can paste this directly into the “Accomplishments / Learnings / What’s next” sections of your Devpost submission.

🏆 Accomplishments that we’re proud of

Autonomous AI Orchestration: The Postman AI Agent plans and executes the full workflow end-to-end, choosing and invoking Ripgrep, Claude, GitHub, and Slack automatically.
Deep Postman Ecosystem Use: Integrated 12+ Postman tools AI Agent Block, Flow Modules, Actions, HTTP blocks, Mock Servers, Monitors, and Analytics.
Multi-API Coordination: Combined 4 external APIs with proper error handling, retries, and dynamic branching.
Smart Conflict Detection: Scans open PRs, compares overlapping files, and assigns a merge-risk score (0–100%).
Transparent Reasoning: Every AI decision, tool call, and prompt is logged in Postman Analytics — complete traceability.
Live Slack Integration: /impact → PR created → team notified — all in real time.
Next.js Dashboard: Displays PR history, conflict hotspots, and risk metrics in a clean, production-ready UI.
30-Second E2E: From Slack command to GitHub PR creation in under 30 seconds.

💡 What we learned

AI Agent > Decision Blocks: One AI block replaced 50+ manual logic branches.
Prompt Engineering is everything: Multiple iterations made Claude’s PR output concise and reliable.
Conflict detection is nuanced: Fuzzy path matching and risk scoring were key.
Slack UX matters: Block Kit formatting made AI notifications actionable.
Reusable Flow Modules: Each tool (Ripgrep, GitHub, Claude, Slack) can plug into future projects.
API resilience: Handling rate limits, retries, and errors across four APIs took careful planning.
Local testing hurts: ngrok’s constant URL changes reinforced the need for true cloud deployment.