AI productivity platform
Montlify
One workspace holding 33+ AI tools for writing, research and chat, with the accounts, subscriptions and infrastructure needed to keep them all running.
- React
- LLM APIs
- Streaming
- Authentication
Overtone
A published Chrome extension that turns an earnings call you are already listening to into research you can defend: answers cited back to a timestamp, slide capture in the player, hedging analysis, and no claim without a source.
Open Chrome Web Store
Product design, extension build and Web Store release
5.0★
Rating on the Chrome Web Store
Cited
Every answer links back to a moment in the call
MV3
Published through Chrome's Manifest V3 review
What made it hard
A summariser produces prose you cannot check, which is close to useless in equity research — an analyst has to be able to defend a claim to someone who will go and listen to the call themselves. So the constraint was set at the start: every statement the tool makes has to point at the second of audio it came from, and anything it cannot source has to be shown as unsourced rather than quietly smoothed over.
How it was built
- 01
Citations are timestamps
Every answer links to the moment in the call it came from, so verifying a claim is a click rather than a re-listen. This is the whole product; the summarisation around it is the easy half.
- 02
Capture inside the player
Slide capture wired into the video controls and stamped with the time it was taken, so a deck screenshot keeps its place in the call rather than becoming an unlabelled image in a folder.
- 03
Hedging analysis
Measuring management's hedging language per thousand words. A quantified read on how guarded an answer was is the kind of signal an analyst forms intuitively and cannot easily evidence.
- 04
Flagging what was not answered
Surfacing the questions that were asked and dodged. The gap in a transcript is frequently the most informative part of it, and it is the part a summariser reliably drops.
- 05
Stripping invented citations
Unsourced numeric claims are surfaced rather than hidden. A tool that occasionally invents a figure is worse than no tool, because a single bad number in a research note costs more than the time the tool saved.
What it does
- Question answering with timestamp citations
- Slide capture in the player
- Timestamped notes
- Hedging analysis
- Unanswered question flagging
- Citation verification
More work
Other things we've shipped.
E-commerce platform
Agento Wear
A production storefront built from nothing: catalogue, guest checkout, payments, order processing, automated email and an admin dashboard the owner runs without an engineer on call.
- Next.js
- React
- Supabase
- Payments
Have something like this in mind?
A short call, then a written scope with a fixed price and a date. Scoping costs nothing and the document is yours either way.