The seven winning projects

Every project links to its Devpost submission (full write-up, images, demo video), its repo, and its live demo wherever one exists. 1st and 2nd place are verified from official sources; the rest are badge-verified winners whose exact slot was never published.

InteractionKit

Harpita and Harpith Pandian · 1st place · $30,000 · verified
What it is

A starter template for gesture-controlled apps: MediaPipe hand, face and head tracking behind a clean Python layer, 17 ready gestures, one setup script and you are building. Two demo apps prove the skeleton: Math-O-Lantern, a pinch-to-slice math game, and Holo-Board, a tool for drawing in mid-air over your webcam and exporting one clip. Kiro's own framing of why it won: solving a specific pain point, the difficulty of vibe coding complex computer vision logic without hallucinations.

Kiro's official spotlight — the gesture layer in motion.
Math-O-Lantern gameplay
Math-O-Lantern: pinch to slice the pumpkin with the right answer.
Holo-Board mid-air drawing
Holo-Board: mid-air drawing over the webcam feed — a content tool, not a game, on the same skeleton.
Under the hood
Math-O-Lantern, 870 linesGameState enum, then Question, Pumpkin, Effect, PumpkinManager and QuestionManager classes, then the game class. A 50-question bank (25 addition, 25 subtraction, four-digit numbers), 10 per session, 15-second timer, 3 lives, 5 points per correct answer, pumpkin spawn, fall and cursor-collision loop. Nine assets including music and sound effects.
Holo-Board, 1883 linesA GestureController with edge detection so each gesture fires once per transition, pen-state priority (rock clears the canvas, pointing draws, thumbs-up cycles an 8-color palette), a transparent drawing canvas composited over the live webcam, PyAudio capture, recording and export.
The kit APINine hand-gesture modules and Hand, Head and Face tracker controllers, each exposing a typed state dataclass. Apps read a single field and never touch computer-vision code.
Their Kiro payloadThree gesture-integration hooks, two complete specs (requirements, design, tasks), five steering docs, Context7 through MCP with auto-approve. Both demo apps were spec-built.
Their actual Kiro files, annotated
The gesture-integration hook.

A hook is a trigger plus a runbook. It fires on new gesture files and walks the agent through seven numbered integration steps, ending with a demand to confirm what gesture was detected and list the files modified. That report-back is the accountability behind the claim that seventeen gestures were integrated in seconds.

The game-development steering doc.

Rules numbered like law: assets first, one test file as the final step, check gesture attribute names before using them, use Context7, respect the template structure, keep detection and game logic separate. Plus a hard rule that application code lives in its own folder and never modifies the core library, and an eight-step workflow ending in documentation.

The Math-O-Lantern spec, requirements and tasks.

Glossary-first: eighteen exact terms defined before any user story, with when-then-shall acceptance criteria. Every task traces to requirement numbers, and one task is a property-based test with named properties: the bank contains only valid operations, and operands stay within digit limits. The champions were doing property-based testing in 2025.

The MCP settings.

Context7 with auto-approve for documentation lookups, credited in their write-up with removing hallucinations on MediaPipe landmark indices. One well-chosen server with auto-approve beats five half-used ones.

  • The 1st-place shape is a Skeleton Crew shape — a lean template plus two demo apps in different genres. The category brief is the winning pattern.
  • Champion DNA: the same duo won the Google Cloud Run 2025 Grand Prize (Storytopia) and second place at Datadog. Browse their repos ↗.

UnitSet

Adithya Reddy · 2nd place · $20,000 · verified on their Devpost page
What it is

A design-to-code platform that replaces six tools: an infinite canvas with drawing tools, wireframe-to-UI generation via AI agents, chat-based creation across four frontier models, vision references from any screenshot, a visual click-to-style editor that writes Tailwind back to source, a code explorer, and Remix the Web — a Chrome extension that captures any element from any live site and reconstructs it as an editable component. Each screen gets its own live Next.js sandbox embedded in the canvas.

Under the hood
StackNext.js 16 and React 19, Convex for real-time sync, E2B per-screen sandboxes, Inngest with AgentKit for the AI workflow, Clerk for auth and billing, Tailwind 4, a custom canvas engine with RAF-throttled reducers.
Kiro payload, the field's largest21 specs covering canvas state, autosave, AI chat streaming, the visual editor, the extension and billing. Fifteen steering docs. Nine hooks, including a component scaffolder on file create, import cleanup on file delete, schema sync on save, and docs auto-update. Four MCP servers, including Chrome DevTools.
  • Their own page says it outright: Kiro became the orchestration engine for the entire development lifecycle. The 2nd-place write-up is a Kiro-usage masterclass — read it end to end.
  • Category-naming works: the Frankenstein of UI creation mirrors the category brief verbatim.
  • Chrome DevTools through MCP let the AI see their UI and self-correct — the experimentation judging box, ticked with a story.

Champa

solo builder · badge-verified winner · slot unpublished
What it is

A unified AI inbox for Gmail, Slack and Calendar. Every message runs a parallel LangGraph agent pipeline: summarization, intent classification, 0-to-10 priority scoring, action-item extraction, spam detection. Smart replies are drafted in your own voice using persona memory from your sent mail, with human approval before anything sends. Query it naturally — urgent emails this week, meetings tomorrow — on a live WebSocket feed.

Under the hood
StackFastAPI, LangGraph and LangChain, PostgreSQL with Qdrant for vector search, React 18 with RetroUI, Auth0, and native Gmail, Slack and Calendar APIs behind a unified token-refresh middleware.
Kiro payloadOne mega-spec: 17 user stories, 85 acceptance criteria, 59 correctness properties mapped to requirements, with tasks traced both ways. Three hooks: a security scan on edit, a code-quality analyzer, and docs sync. Context7 and LangChain docs through MCP. Hypothesis property tests at 100+ iterations each. Over 10,000 lines, solo.
  • 59 correctness properties bridging spec to executable tests is the most-quoted number in their build story. Numbers like this are what thorough looks like to judges.
  • A security-review hook that caught several near-misses with committed credentials — a hook with a saved-my-bacon story beats a hook that runs lint.
  • Solo, unsexy productivity still badged. Depth of workflow beats team size or theme.

Strangler Studio with PHP Migration Tool

badge-verified winner · slot unpublished
What it is

An AI-powered migration tool for teams stuck on legacy PHP. Pitch: stuck with legacy PHP code holding your team back? Meet Frankenstein Laboratory. The direct project page resisted our extractors; the gallery card opens fine from a normal browser.

  • Developer tools took three of seven badges — InteractionKit, UnitSet, Strangler. Build for builders.

GeoSpirits

Ak Deepankar and team · badge-verified winner · slot unpublished
What it is

GeoCities reborn, but every page you build is pinned to a real place on an interactive 3D Mapbox globe. A drag-and-drop builder with ten-plus components, AI-generated pages and layouts, embeddable AI chatbots per page, shareable URLs, a My Sites dashboard, and a spooky mode with flying ghosts. The same democratized web publishing that made GeoCities magical — now that corner has coordinates.

Under the hood
StackNext.js 16 and React 19, Supabase for Postgres, auth and row-level security, Mapbox GL globe projection with reverse geocoding, OpenAI page generation, Vitest with fast-check property-based testing at 100+ iterations.
The hybrid rule, their honest numbersVibe-coded the AI generator first: a 30-minute prototype with no validation. Rewrote as a spec: one hour speccing plus four hours of Kiro implementation, production-ready with property tests. Their rule: vibe for UI and simplicity, spec the moment external APIs, state and security enter.
  • The vibe-versus-spec comparison with real time numbers answers the judges' write-up question — how spec-driven development compared to vibe coding — perfectly. Steal the format.
  • Two words won a badge: Resurrecting GeoCities. The clearest category read of 2025.

DebateMaster-AI

badge-verified winner · slot unpublished
What it is

A real-time voice debate coach: bidirectional audio streaming with the Gemini Live API, an adaptive AI opponent, instant post-debate feedback and progress tracking. Two modes, Coach and Fierce, a WebGL voice-reactive visualizer, and an analytics suite — confidence trends, duration distributions, activity heat maps — plus a persona and archetype system.

Under the hood
StackReact with Vite and Tailwind, the Web Audio API, the Google generative-ai SDK, and dashboard components for heat maps, trends and persona showcase.
Kiro payloadFour specs: the core app (15 user stories, 30+ correctness properties, 33 tasks with property-test specs), persona showcase, analytics dashboard, and the live-debate UI redesign. Steering docs for product vision, structure and conventions.
  • Analytics is the proof of value: the coach claim is backed by trend charts and heat maps — performance analytics as product surface, not a marketing line.
  • Repo-match honesty: the README confirms it is powered by the Kiro IDE and Gemini Live with a dedicated how-we-leveraged-Kiro section, though it never names Kiroween, so this stays at strong-match confidence.

Soulful Sessions

Ansell Maximilian · badge-verified winner · live on the Chrome Web Store
What it is

A Pomodoro-RPG Chrome extension: focus sessions damage Stubborn Souls — bosses with backstories about procrastination and unfinished work, each ending in a heartwarming resolution conversation when defeated. The dopamine-detox twist: game UI is hidden while you work and unlocks only during breaks. Work, then reward. Soft-discouraged versus strict-mode hard-blocked sites, ten bosses, three stats, cosmetics, idle collection, task management, streak stats.

Under the hood
StackManifest V3, TypeScript in strict mode, vanilla JS and CSS with no frameworks, esbuild, declarativeNetRequest blocking, Jest with Puppeteer, a background service-worker game engine.
Kiro payloadOne spec of 42 tasks, executed nearly hands-off — the builder writes that he barely touched the code himself. A task-completion hook auto-commits on every finished spec task, giving a full build history. Later specs added the boss bestiary, a player card, and a dev screen to test all game states. Zero to one hundred purely with Kiro.
  • The shipped-store lesson: approved on the Chrome Web Store during judging — the working-app requirement satisfied by an install link anyone can click. If our entry can ship to a store, that is the strongest possible demo.
  • A calming aesthetic beat aggressive productivity — design taste as differentiation, no flashy tech needed.
  • Extensions are underserved — one of seven badges went to a zero-framework Manifest V3 extension.

2026 rival watch

seance-relay-chat

dyc19991225 · Resurrection — a real IRC server (RFC 1459), web client and AI Ghost bot. Honesty-forward Kiro log.

seance

hassaanraja282-lgtm · Resurrection — summon dead web pages via the Wayback Machine.

gravekeeper

mangeshraut712 · resurrect abandoned repos and dead pages via Wayback, with hash-chained receipts.

necromancer, plus crew

toriyama237 · Frankenstein — byte-for-byte failure replay across COBOL, FTP, Postgres and Rust, with a four-agent Kiro Crew companion.

dormbit

AmirmLotfy · Tauri and Rust, local-first dormant-project preservation on macOS.

ghostlist

liuwei412815 · a playful recurring-task manager with Kiro-built scheduling.

The gap

Nobody visible competes on art or design strength, and only one extension is in the field. The Costume student lane stays open.

How the winners used Kiro

InteractionKit, 1st3 hooks · 2 specs · 5 steering docs · 1 MCP server · property-test tasks inside the specs
UnitSet, 2nd21 specs · 15 steering docs · 9 hooks · 4 MCP servers including Chrome DevTools
Champaone spec with 17 stories, 85 criteria, 59 properties · 3 hooks · 2 MCP servers · Hypothesis property tests at 100+ iterations
GeoSpirits2 specs · a hybrid vibe-spec workflow with time numbers · fast-check property tests at 100+ iterations
DebateMaster-AI4 specs · 15 stories · 30+ correctness properties · 33 tasks with property tests
Soulful Sessionsa 42-task spec · a task-completion auto-commit hook · a dev-screen spec · built zero to one hundred hands-off

The pattern: every single winner paired specs with property-based tests and stated quantities — stories, criteria, properties, tasks. The 2026 judging criteria ask about property-based testing explicitly, and the field was already answering in 2025.

Method

The Devpost winners tab is Cloudflare-walled with zero archive captures, but direct submission-page URLs and the paginated project gallery extract cleanly. Gallery Winner badges — seven in total, all on the first three pages — reconstructed the board. 1st place is verified via the repo and Kiro's official spotlight; 2nd place is verified on UnitSet's own Devpost page. Strangler's direct page resisted extraction, so it links to its gallery card, which may open fine in your browser.

Confidence labels: verified means an official source names the placement. Badge-verified means the gallery badge is present but the slot was never published. Strong match means repo evidence matches the gallery text, as with DebateMaster. The full source log lives in the notes alongside this site.