404NOTmissing

Inspiration

During the opening presentation of HackEurope here in Paris, we experienced something small but revealing: we were told that every participant should have receive a hackathon bracelet. But the bracelets never arrived.

It wasn’t a catastrophic failure, but it exposed a deeper issue. In complex, fast-moving organizations, decisions are made in meetings, tasks are verbally assigned, but execution often fails.

We started asking:

Was the budget approved?
Did the Finance department notify the Operations department?
Who placed the vendor order?
Was shipment verified?

Most likely, all of these steps were discussed. But discussion is not enough, there needs to be coordinated action after all the meetings. Teams don’t have a persistent, structured memory of what was decided, who owns it, and what depends on it.

So we built a Meeting Memory Layer.

What it does

404NOTMissing transforms raw meeting transcripts into structured, executable actions.

Instead of meetings producing passive notes, our system:

Ingests meeting transcripts
Extracts structured actions
Assigns them to departments or individuals based on the organisation schema of the company
Tracks urgency and status
Creates a live execution board

Every decision becomes:

An action (e.g. order 300 bracelets for HackEurope Paris)
- With an owner (e.g. Lead of Operations department)
- With a deadline (e.g. 3 business days)
- With a status (e.g. TO DO)
- With escalation if needed

The result:

No lost tasks
No ambiguous ownership
No invisible dependencies
No “Did we do that?” meetings

The meeting transcripts stop being documentation, they become an execution engine.

How we built it

We designed 404NOTmissing as an agent-driven execution system.

Transcript Ingestion We start with meeting transcripts (synthetic for this demo), representing realistic multi-department coordination inside HackEurope. Each transcript is stored as raw text in our database.
Agent-Based Extraction Layer Instead of building a full knowledge graph infrastructure, we implemented a set of focused AI agents that operate over transcripts to extract structured execution data. Our agents perform:
Entity extraction (people, departments, projects)
Task detection (actionable commitments vs discussion)
Sequence identification (what depends on what)
Priority inference
Ownership assignment
Structured Execution Model (Database-Driven) Extracted actions are stored in a structured database schema, which acts like as a lightweight “meeting memory layer” — not a full knowledge graph yet, but a persistent execution record.
Execution & Integrations Once structured, actions can trigger operational workflows:
Ticket creation (in Linear)
Phone call reminders (via ElevenLabs)
Email dispatch