🤓 Latest Submissions
DevPilot: AI Developer Workflow Copilot
The problem: every engineering team runs the same five workflows -- onboarding, code review, testing, release sign-off, and legacy modernization -- and they stay manual and disconnected. A new engineer reads unfamiliar code with no map. Reviewers re-derive the same risk checklist by eye on every diff. Coverage gaps surface after an incident, not before one ships. Release sign-off is a checklist that's easy to skip under deadline pressure. Tech debt piles up with no systematic way to triage it. DevPilot is a multi-agent copilot that answers all five at once. One orchestrator dispatches six specialist subagents in parallel (onboarding, code review, testing, release readiness, modernization, and auto-fix) and reduces their output into a single, role-scoped, decision-ready report -- instead of a human running five separate tools and stitching results together by hand. Findings are cross-checked, not just flagged: a static finding is promoted to CONFIRMED only when an independently generated test actually fails against the running app. On our seeded demo (TaskFlow API), two findings were confirmed this way with zero false positives; the same pipeline was then proven cold against a real, unmodified public GitHub repo (a Flask app it had never seen) to show the scanners aren't hardcoded to one framework. Beyond the five core workflows, DevPilot adds: Blast Radius (a real reverse-dependency search over the repo's own import graph -- "if this breaks, here's exactly who else feels it," not a simulation); Pre-Mortem Intelligence (a plain-English failure timeline built from that same data); Executive Intelligence (health score, release call, hours saved -- deliberately refusing to invent MTTR/MTBF numbers a static tool has no telemetry to back); an Engineering Historian (a local run-history log surfacing score trends and recurring weaknesses across runs); and a full Dependency Map of the repo. Target users are engineering teams and individual developers who want a pre-merge safety net: new hires get an instant architecture map, reviewers get severity-ranked, corroborated findings, testers get real coverage gaps plus auto-generated tests, release managers get a cited GO/NO-GO call, and leadership gets one honest page. Users interact through a live web app: paste a GitHub URL or a local path, watch six subagents report in as they finish, then explore Ask DevPilot (grounded Q&A over the real report) and "Make it flawless" (safe, mechanical fixes with a diff preview, gated behind explicit human confirmation, committed locally -- never auto-pushed). What makes it creative and unique is the honesty built into the design: every number traces to a real scan or a real test run, corroboration is measured not asserted, and features that would need data DevPilot doesn't have (fabricated incident telemetry, invented reliability metrics) are explicitly declined rather than faked. It's proven twice -- once on a seeded demo with known bugs, once cold on a real, unrelated open-source repository -- rather than tuned to look good on one fixture. The result: five reactive, manual workflows become one automatic, pre-merge, cross-checked pass a team can actually trust.
27 Sep 2026
