AI

DependencyWatch

Catch the signal before the break — real-time dependency health scanner.

🏆 Hackathon Winner🤖 AI PoweredHackathon Project
DependencyWatch

Overview

An agentic dependency-health platform that scans a project manifest (or any public GitHub repo) and fuses four independent signals — security CVEs, maintenance activity, adoption, and live web chatter — into a single 0–100 'repair signal' per package, with a recommended action and a citable evidence trail. Built on Clean Architecture so every external provider degrades gracefully. Winner of the Tower Pipeline Challenge at the DeveloperWeek New York 2026 Hackathon.

Key Features

  • Four-pillar scoring — Security (45%), Maintenance (30%), Live Chatter (15%), Adoption (10%) — fused into one repair signal
  • Security-first override that caps vulnerable packages in the red, matching npm-audit behavior
  • Multi-ecosystem scanning: npm, PyPI, Go, Rust, Maven, and SBOM (CycloneDX/SPDX)
  • AI fusion chain: Claude → Groq → deterministic engine, with graceful degradation
  • Live results streaming via Server-Sent Events, worst dependencies first
  • Tower Python pipeline with parallel fan-out and an Apache Iceberg lakehouse cache for instant re-scans
  • CI/CD GitHub Action that fails any PR introducing a critical dependency
  • PDF/JSON report export, saved projects, and auto-generated alerts