Speak Meetings: AI in production, five years early.

AI meeting intelligence in production in 2018 — Microsoft and Google shipped equivalent features years later. This is what building AI products before the LLM era looked like.

Context

Meeting intelligence before it had a name

In 2018 there was no commodity model you could call to transcribe a meeting, pull out the action items, and route them to the right people. If you wanted meeting intelligence, you built it: transcription, natural-language processing, and coordination logic, assembled from custom pipelines and made reliable enough for live use.

That is what Speak Meetings was — an AI-powered meeting intelligence platform, in production, years before the platforms made the category familiar.

Constraint

No off-the-shelf anything

Every capability the product needed — speech-to-text quality, entity and action-item extraction, summaries a team would actually trust — had to be engineered from the NLP and ML tooling of the day and hardened against real meeting audio. Production AI without commodity models meant owning the whole pipeline: accuracy, latency, failure modes, and the product judgment about what to show users when the model was unsure.

What We Built

The full loop, not a demo

  • check_circle Automatic transcription — live meeting audio into usable text.
  • check_circle Action item extraction — custom NLP pipelines identifying commitments and owners.
  • check_circle Team coordination — meeting output routed into the follow-up work it implied.
  • check_circle Production operation — a running service with real users, not a research prototype.
What Happened

The market caught up; the IP lives on

The major platforms shipped equivalents years later. When commodity LLMs reshaped the market, the company was wound down and Date Palm Media acquired the IP to preserve it.

The experience did not wind down with it. Building production AI before LLMs is why we treat models as components to be engineered around — not magic — and it is the lineage behind our delivery work and our AI-native builds today.

Stack

Pre-LLM production AI

PythonNLP / MLReactPostgreSQL
Verify It

Check the claims yourself

  • link Archived 2018 product page for the platform’s assistant, via the Internet Archive: alfredbyspeak.com, October 2018 snapshot.
  • link Microsoft Teams intelligent recap and Google Meet AI notes launched publicly in 2023–2024 — public product timelines.
  • link Company fast facts on our press page; founder track record at josephdattilo.com.

We were early once. Now the tools are better.

If you are putting AI into a real workflow and need it to behave like production software, that is the job we have been doing since 2018.

Start a Project arrow_forward