Aura Graph: multi-agent AI knowledge graph platform

Case study: a multi-agent AI platform that turns meeting transcripts and project docs into an interactive knowledge graph of decisions, tasks and risks.

About the client

Confidential client.

The project

A multi-agent, AI-powered full-stack platform that turns meeting transcripts and project docs into an interactive, explorable knowledge graph — surfacing decisions, action items, open questions, risks, and requirements as a force-directed graph, with dedicated views for tasks, questions, requirements, dependencies, and a session timeline.

Type: AI Solution + Web App. Built in 25 days, live at auragraph.infnai.in. Tech: TanStack Start, React, TypeScript, Supabase, Tailwind CSS, shadcn/ui, Gemini, Groq, Mistral.

Key features

• Multi-Agent AI Extraction: A multi-phase extraction pipeline spanning Gemini, Groq, and Mistral with per-phase fallback routing — reads uploaded transcripts and docs, and proposes structured graph content phase by phase
• Interactive Knowledge Graph: A force-directed, explorable graph of decisions, tasks, risks, requirements, and more, with dedicated Kanban, dependency-chain, and timeline views layered on top
• Secure Auth & Collaboration: Email/password and GitHub OAuth authentication, workspace/project RBAC, and built-in collaboration — comments, assignments, @mentions, and a review queue for every AI-proposed change

Results

Aura Graph gives teams a single, explorable source of truth for what happened across meetings and project docs — replacing scattered transcripts and notes with a living knowledge graph that surfaces decisions, risks, and open work automatically. The multi-agent AI pipeline turns raw uploads into structured, reviewable graph content in minutes instead of hours of manual note-mining, while the human-approval step keeps a team fully in control of what becomes part of their project's record. The entire platform — graph engine, multi-provider AI pipeline, authentication, RBAC, and collaboration tooling — was scoped, built, and shipped in 25 days.

Honestly we didn't expect to get a working product this fast. We're talking a full AI pipeline, a real knowledge graph, logins, permissions, the whole thing, done in 25 days. What matters more to us is that it actually works the way we use it. Stuff that used to get buried in meeting notes just shows up now, and we still get to approve what goes into the graph before it's final, so nothing feels out of our hands. It's probably the smoothest build we've had with any dev team so far.