HackMate AI Hackathon: M Your AI teammate for surviving a hackathon. Idea & Problem-Solution Document LH × EM K Center × UniDevs × M L Bangladesh · August 17, 2026 Tracks: Best Use of Gemini API · ElevenLabs Side Track
- Industry Hackathons / Developer Productivity / AI Tools. Hackathon participation has exploded in Bangladesh over the last three years. M LH-style events, university-organized sprints, and M L Bangladesh community jams now draw thousands of undergraduate builders every year. Yet the toolchain around them has barely changed. Today, a hackathon participant uses: A static Trello board (or nothing at all) A group chat where someone yells "30 minutes left!" twice A mental TODO list that drifts the moment an API call fails A last-minute scramble to remember what Devpost requires The industry is real, growing, and underserved.
- Problem Hackathon participants struggle to effectively manage limited development time, and frequently lose time on low-priority tasks — leaving insu fficient time for testing, demos, and submission. Specifically, every participant faces the same compounding pressures simultaneously:
- Idea selection — what to build, how to scope it
- Requirements research — what judges will reward, what platforms expect
- MVP construction — actually shipping a working core
- API integration — wiring Gemini, ElevenLabs, etc.
- Debugging — inevitable, expensive, demoralizing
- Test runs — multiple devices, multiple accounts, edge cases
- Demo prep — script, storyboard, recording
- Demo video — under 3 minutes, often re-recorded three times
- Devpost writeup — name, description, tools, prize tracks, screenshots
- Submission — before the literal deadline The problem isn't simply time management. The problem is decision-making under time pressure: knowing what to work on right now and — harder — knowing what to sacrifice when reality changes. The pain is observable in every event: Demos that fall apart because the build was never finished Devpost submissions with broken screenshots or missing tools Teams that built a beautif ul secondary feature but never had the demo Solo builders who spent 4 hours on polish and submitted 12 minutes late There is no existing tool purpose-built for this. Notion templates are static. Trello boards don't replan. Calendar reminders say "1 hour left" but don't say wh a t to do.
- Solution HackMate AI is a web-based AI teammate that monitors your hackathon state and continuously produces an adaptive execution plan. You tell HackMate: ⏰ Your hackathon deadline 💡 Your project idea 👤 Team size (solo or team) 🛠 Your current progress 🎯 Your target prize tracks 📋 Your submission requirements 🚧 Your current blockers (if any) HackMate uses Gemini to generate a prioritized plan that adapts every time reality changes. What makes HackMate different from a chatbot Normal AI assistant HackMate AI "Here are some things you could do." "Here's what you should do next." One-shot answer Continuous state tracking Static advice Dynamically replans when reality changes Normal AI assistant HackMate AI Generic productivity Built specifically for deadline-driven building Doesn't know about Devpost Tracks submission readiness in real time Core capabilities Initial Planning — Gemini ingests project context, deadlines, prize tracks, and progress to produce a ranked execution plan with explicit time budgets per task. Replanning — The centerpiece. Whenever a blocker appears or a task runs over, HackMate asks Gemini to recalculate the plan: what to drop, what to keep, what's the minimum viable submission. This is what every participant wishes they had access to. Submission Guardian — A live checklist (working app, demo video, GitHub repo, Devpost writeup, screenshots, tools, prize tracks) with a submission-health percentage that updates as you complete items. Voice Coach (ElevenLabs) — A one-tap spoken briefing so the user can hear their current situation hands-free, usef ul for the long final stretch. Coaching Insight — Gemini's reasoning surfaces as a coach message in plain language: "Don't add a leaderboard. You have 58 minutes left and your demo isn't recorded."
- Why this solves a real problem This is not an abstract productivity tool. We are using HackMate while building HackMate. Every member of the team has felt this exact pain at every hackathon we've attended. The product is a direct response to lived experience. The validation is immediate: Hackathon organizers want higher-quality submissions — HackMate helps every participant submit something complete. Individual participants want to maximize their judging chances — HackMate forces the right trade-offs. Judges benefit because they actually get working demos to evaluate. A f uture version could expand to: Team-mode with shared state across multiple builders Integration with submission-platform APIs (Devpost, M LH) for auto-fill Historical replan analysis ("teams that replanned within 60 minutes of a blocker submitted 2× more often") Time-vs-judging-score modeling to recommend which prize track is most winnable
- Innovation / Why it's impressive Most "AI hackathon projects" are wrappers over a single API call. HackMate is meaningf ully different for three reasons:
- Statef ul planning, not chat. The product remembers your project across the entire session and treats every interaction as input to the next decision.
- Replanning under uncertainty. When the user says "lost 25 minutes debugging the Gemini integration," HackMate asks the model to remove items and rebalance the schedule — the same reasoning an experienced hackathon mentor does, but available instantly and repeatedly.
- Voice as a first-class surface. ElevenLabs isn't tacked on as a "play my text" button. The voice briefing is an alternative input/output mode for a coach relationship — closer to a personal trainer than a chatbot. The combination is rare: state + replanning + voice, applied to a setting where every second matters.
- Target users Primary: Solo or small-team hackathon participants at undergraduate events in Bangladesh (M LH, UniDevs, M L Bangladesh, EM K Center, BRAC University, BUET, NSU, IUT, etc.) who need adaptive guidance during an 8-hour or weekend hackathon. Secondary: Hackathon mentors / organizers wanting participants to submit more complete entries. Bootcamp students participating in timed build challenges. Indie hackers running personal 48-hour sprints who want the same adaptive structure.
- Technology Layer ChoiceReason Frontend Plain HTM JS L + CSS + No build step, fastest to ship, easy to host statically AI brain Google Gemini API Strong structured-JSON reasoning, fast gemini-1.5-flash model fits replanning latency budget Voice ElevenLabs API eleven _ multilingual v2 _ English/Bangla briefings voice, natural sounding Layer ChoiceReason State Browser localStorage Hosting Netlify / Vercel static Zero-backend MVP, keys persist on the user's machine Free, instant deploy, public U R L in minutes Gemini architecture (brief) A single prompt template that includes: Deadline and time remaining Project name, idea, prize tracks Current progress + completed tasks + remaining tasks Latest blocker (if any) Gemini returns: { "current task": " ... " _ , "current task _ _ minutes": 18, "current task reason": " "tasks to _ _ drop": [" ... "], "submission _ warnings": [" " ... _ _ "up_ next": [{"task": " " ... , ... "coach _ message": " " ... , "minutes": 25}], "], } Replanning uses the same template plus the trigger reason. The model is asked explicitly to remove lower-priority items when time pressure is high. ElevenLabs architecture (brief) Coach message string → POST to /v1/text-to-speech/{voice _ id} → audio blob → play in element. No streaming needed; briefings are 10–25 seconds.
- Demo narrative (90 seconds) The demo doesn't show a happy path — it shows a crisis: Time On screen 0:00–0:15 0:15–0:30 Problem statement: time evaporates during hackathons Setup form: project name, deadline 4:00 PM, progress 40% Time On screen 0:30– 0:45 Generate Plan → dashboard appears with current mission 0:45–1:05 "I'm blocked" → "Gemini integration failed, lost 20 min" → replan drops low-priority 1:05–1:20 1:20–1:30 1:30–1:45 items Voice Briefing button → ElevenLabs speaks the coach message Submission checklist ticks update health percentage Closing: "HackMate turns your remaining time into the next action" This narrative is what judges remember: a planner that adapts in front of them, not a chatbot that answers a question.
- Scope discipline (what we are explicitly NOT building) To protect the build window: ❌ Authentication / accounts ❌ Database / multi-user persistence ❌ Team collaboration features ❌ Real-time multiplayer ❌ GitHub integration ❌ Slack/Discord integration ❌ "Should I build this?" feature advisor ❌ Mobile-native apps ❌ Complex analytics Every dropped feature reduces risk of having nothing ship-ready at submission time.
- Future vision If HackMate reaches users, the natural next steps are: Team mode — shared state across multiple builders, conflict-aware replans Submission-platform integrations — Devpost and M LH API auto-fill Multi-language briefings — f ull Bangla and Hindi ElevenLabs voices for South Asian hackathons Historical analytics — which replanning patterns correlate with on-time submission Antigravity-built native UI — desktop app that runs in the system tray during long hackathons
- Why this deserves to win Best Use of Gemini API: Gemini is not a chatbot here — it is the planning brain, responsible for structured-JSON scheduling, replanning under blockers, and generating the coaching insight that drives the whole product. ElevenLabs Side Track: Voice is not a gimmick here — the spoken briefing is the user's hands- free mode during the final hour when they're typing f uriously and need to know what comes next. Demo: The replanning sequence is a single concept any judge can grasp in 90 seconds: "watch the AI cut scope when this team loses 20 minutes." Real problem: Solved by the builders, for the builders, validated by the act of building it during the event itself. End o f do cument.
Log in or sign up for Devpost to join the conversation.