Insights

What you submitted

Multi-agent AI system that intercepts GitLab PRs, checks them against real Dynatrace production incident history, fans out across every service in the codebase in parallel, and autonomously opens fix MRs — each citing the specific incident that proved the pattern was dangerous. Demo: a 12-service codebase where incident P-26053 (a 47-minute, £23k outage) shows the same HTTP timeout pattern in 8 of 12 services. Four FastAPI microservices on Cloud Run (Orchestrator → Intelligence → Scanner → Fix Factory) with a real-time Next.js WebSocket dashboard. Real Dynatrace integration, MongoDB institutional memory, ADK agents, 176 tests, TDD throughout.

What you submitted

Real-time room availability and flow-tracking for university buildings. ESP32 microcontrollers with ultrasonic sensors at doorways detect direction of travel (entry vs exit), feeding live headcounts to a web dashboard. Students see occupancy, seats remaining, whether a room is filling or emptying (rate-of-change), historical patterns, free-room notifications, and 'Hidden Gems' (quieter nearby spaces), plus natural-language queries via Gemini. Staff monitor bottlenecks and evacuations. Custom 3D-printed sensor enclosure and Arduino firmware. Stack: ESP32, AWS Lambda (Python), Vultr Cloud Compute analytics engine, MongoDB Atlas, Google Gemini, Next.js 16. Built by 4 people in 24 hours.

What you submitted

Serverless AI-powered email triage. Users get a dedicated proxy email address (@inboxpilot.xyz); incoming mail is auto-classified by Claude 3 Haiku via AWS Bedrock into categories (Sales, Applications, Spam, Partnerships, etc.). Users view categorised tabs, flag urgent messages, ask natural-language questions ('all emails from john@example.com about a missing payment'), and configure per-category auto-reply templates via SES. Fully serverless: three Lambdas (Authorizer, API, Triage) + API Gateway + DynamoDB + S3 + SES, with a Next.js 15 frontend. IAM roles scoped to least privilege.

What you submitted

Decentralised community crime reporting platform with a Twitter-card UI. Users submit incident reports with media, pinned to a Google Maps layer. Verified reports earn 'Krime Koins' — Verdn-powered tokens to incentivise civic engagement. An AI chatbot called 'Peter' answers safety questions and detects SOS keywords to trigger emergency UI flows, with voice recognition listening for 'help me', 'emergency', or 'peter peter'. Built full-stack in 24 hours by a team of 2. Stack: React 18 + Vite, Tailwind/DaisyUI, Firebase Auth, Google Maps, Node/Express, MongoDB, Zod, Google Gemini 1.5 Flash 8B, AWS S3, Verdn API.