Turns meetings into tasks that get done
A fast-growing tech company was drowning in meetings, losing track of decisions, and spending hours creating tasks by hand. We built a multi-agent Slack chatbot that searches meeting history with RAG, extracts action items, and syncs them into the team's project tool.
Every reply carries its source.
3
specialised agents behind one bot
1
channel your team already lives in
0
answers given without the meeting they came from
Decisions were being made, then lost
The meeting happens, the decision gets made, and then it lives in a recording nobody re-watches. The work after the meeting was where the team's time went.
Before
- Long post-meeting work, creating and assigning tasks by hand
- Lost meeting context — important decisions buried in recordings
- Action items getting forgotten between one meeting and the next
- Disconnected tools: Slack, the call recorder and the task board not talking
After
- Tasks created and assigned from the decision itself
- Any decision findable by asking a question in the channel
- Action items tracked rather than remembered
- One workflow: the tools stay, the syncing stops being somebody's job
Three specialised agents, working together
One agent that talks, one that finds, one that files. Splitting them means each can be tested, replaced and reasoned about on its own.
Conversation agent
Handles natural language interactions in Slack, understanding context and intent — intent recognition, context management and multi-turn dialogue.
RAG search agent
Performs semantic search across all indexed meeting transcripts — vector similarity, context retrieval and result ranking.
Task automation agent
Extracts action items and creates tasks in project management tools — action extraction, task prioritisation and assignment.
A decision you can check
Ask a question in the channel and the answer comes back with the meeting, document or ticket it came from, so the decision can be checked rather than taken on trust.
That matters most when the answer is inconvenient. A bot that cannot show its source is a bot the team stops believing the first time it is wrong.
Built with
Bring the meeting nobody wrote up
This is our AI engineering practice
It is real work and it is where our four products came from. But what Cognilium leads with is narrower: optimization apps that run in tandem with Microsoft Dynamics 365, computing the decisions the ERP records but does not derive — the optimal price, the optimal pick path, the optimal stock level. See the optimization apps · How we build inside the ERP.