HackaLakaBoomBoom

Landing Page
Login Page
User dashboard
Audio Briefing
Flowchart
Deployment

Inspiration

We talked to people working across engineering, product, marketing, and leadership. The same story kept coming up: decisions live in meetings, Discord threads, and docs, but nobody shares one version of the truth. We wanted AI that doesn’t just summarize noise; it verifies, flags contradictions, and speaks each role’s language.

What it does

SyncSpace is a role-aware workspace for team knowledge. You pick a project, process sources (transcripts, chat logs, uploads), and get one cached Insight Store. From that store you get:

Grounded chat with citations that jump to the exact source line Alignment Radar that surfaces contradictions with confidence Role-filtered text & audio briefings (eng / marketing / product / exec) Decision flow (Mermaid) and a Sources pane with provenance Different logins see the same facts framed for their altitude — chat history stays private per user.

How we built it

Next.js three-pane UI (project sidebar · chat · studio rail) Mistral end-to-end: Embed (RAG boundary), Large 3 structured extraction, Small framing pass, Voxtral TTS Insight Store as the single cache — extract once, re-render everything else Supabase Postgres + Drizzle for users, projects, and per-user chat turns bcrypt auth + signed session cookies; deployed on Vercel Live: https://syncspace-peach.vercel.app/

Challenges we ran into

Keeping answers grounded (no invented dates) while still role-aware Making uploads actually enter the model — had to invalidate processed so re-process picks up new files Moving from in-memory demo auth to real DB without breaking the hackathon demo Supabase’s direct host is IPv6-only — needed the pooler connection for local + Vercel Port conflicts and lockfile/deps issues when teammates ran the app fresh

Accomplishments that we're proud of

A full product loop: login → process → radar → cited chat → role switch → decision flow Alignment Radar catching the Sept 15 vs Q4 contradiction in a realistic cross-team seed Production-shaped auth.

What we learned

Cache the truth, don’t re-prompt it. one Insight Store beats five separate LLM calls Role-aware AI is useless without provenance people can click Demo UX and production auth collide fast; ship the DB earlier than you think “Works on my machine” dies on DNS, ports, and package-lock — document the path

What's next for HackaLakaBoomBoom

What's next for HackaLakaBoomBoom Live Slack/Discord connectors (not just seeded logs) Forgot-password / invite flow with email Stronger PDF/DOCX extraction and auto-process on upload Org-level workspaces and shared vs private project permissions Deeper Alignment Radar: suggest reconciliations, not just flags