Fulkram
Fulkram / work / thirty-six million words
The problem: knowledge buried in recordings nobody can search

The answer's in there somewhere.

Six years of my own teaching on video, and every time I needed one specific answer, someone had to scrub through recordings by hand. So I indexed all of it, on my own machine. The answer now takes seconds. Nothing left the building.

The estate

What's in it

Every number on this page is a chip from the ledger. Click one for the method and the date it was verified.

Recordings catalogued

Teaching sessions, coaching calls and workshops going back six years.

Transcripts completed

Transcribed on my own hardware. No per-minute API bill.

Words indexed

The whole corpus, searchable from one place.

Searchable chunks

Ask a question in plain English, get the exact passage back.

Timestamped segments

Every result points at the second it was said, so you can jump straight to that moment in the recording.

Nothing left the machine

Transcription, indexing and search all run on my own hardware. The system never calls an API, so there's no bill to watch and no data-sharing question to answer. The recordings sit on a machine I own, and the index sits next to them.

If you're in a regulated business, that's the whole point. Patient calls, client files, legal recordings: the same build works when the content isn't allowed to leave your building. The system lives where your data already sits. When I hand it over, you own the whole thing.

The knowledge base dashboard: 5,031 transcripts, 36.8 million words indexed, and the pain-point distribution with labels withheld.
Screenshot The dashboard on the real system. Transcript, word and pain-point counts, read live from the database. The bar chart is the 8 customer pain points it surfaced.
Outputs

What it produced

The count is the receipt. This is what it bought.
01

The pain points that rewrote the marketing

Mining the archive surfaced 8 customer pain points, in customers' own words, that no brainstorm had found. They now drive the copy on my own sites.

02

Blog posts, generated and live Live

72 posts on the live blog, each written from a recording. Go and have a look: dylanayaloo.com/blog.

03

A tone-of-voice guide with nothing leaked

Mined from six years of teaching, so anything written in my name sounds like me. Not one member name in it. I ran a machine search for every member's name before it went anywhere near another tool.

  • The blog the corpus writes. 72 live posts. Liveopen the blog
  • The same shape on business documents. Invented data, labelled as such. Prototypeopen the demo
  • Every number above. Method and date for each./claims/

Your version of this

Yours is probably the sales calls, the site surveys, the handover threads, the process that only exists in the head of whoever's been there longest. I don't know exactly what it is in your business without having a look. The build is the same either way: transcribe, index, search, locally if your data demands it. The answer stops depending on who you can grab in the corridor. If that's your week, the door for it is buried knowledge.

One honest caveat. This runs for one operator, me, on one machine. It hasn't served thousands of concurrent users and I won't pretend otherwise. At that scale the architecture changes, and I'd tell you so before anything was built.

If your answers are buried in recordings, documents and old threads, that's a build I've already done once.
Tell me where they're buried