All work

Albert N. Martin Sermon Archive

A 46-year preaching ministry given a permanent home: 2,967 sermons transcribed, understood and made searchable with AI.

Client
Albert N. Martin Sermon Archive · in honor of Pastor Martin, 1934–2026
Location
Montville, New Jersey
What we built
AI archival · full-stack build · preservation
Stack
Astro 6 · Whisper · Gemini · pgvector
The Albert N. Martin Sermon Archive home page, with a portrait of Pastor Martin beside the archive search.
The archive home page on a phone.

A home for a life’s work

Pastor Albert N. Martin preached at Trinity Baptist Church in Montville, New Jersey, for forty-six years. When he passed away in April 2026, nearly three thousand sermons existed only as audio: no transcripts, no index, no way to find a sentence without listening from the beginning.

The work had two callings: a worthy, permanent home, because this archive is a memorial as much as a library, and access, turning more than 2,500 hours of the spoken word into something a person can search, read and share.

“Not unto us, O Lord, not unto us, but unto thy name give glory.”

Psalm 115:1, the verse that greets every visitor to the archive
  • 2,967 sermons preserved across 225 series
  • 46 years of ministry one pulpit, one man
  • 26,452 illustrations indexed stories, quotes, analogies

Source: the live archive and its source project, September 2026.

From a phrase to his voice

Most sermon sites let you search titles. This one searches every word he preached, and every result leads back to the moment he said it. Choose a phrase to search.

Try a phrase

Matched in a sermon transcript

  1. Transcript · Play from 67:18 Puritan Evangelism “When I have shot and spent all my gospel bullets and have none left and little effect seems to be made upon my hearers, I then put myself in the gun and shoot myself at them.”

Illustration of the archive’s search; the results and screens are from the live archive.

The live archive’s transcript of Puritan Evangelism: three quotations in italics, each with a Play from button (48:49, 53:47, 67:18) and a Share link.
The live archive: the line in the transcript of Puritan Evangelism, with Play from 67:18.

A single sermon page, Puritan Evangelism, showing a written summary, a “Listen Now” button, and “52 illustrations in this sermon”.
Each sermon page carries a plain summary, its topics and Scripture, and the illustrations inside it.
Browsing sermons by Scripture: every chapter of Romans with its associated sermons.
Or start from Scripture: every chapter of a book, with the sermons that expound it.

Teaching a machine to listen

We ran the whole corpus through Whisper, a speech-recognition model, so every paragraph has a timestamp; that is what makes each line click-to-play. Then Google Gemini read every sermon and pulled out its structure: summary, topics, the passages it expounds, an outline, application points and every illustration he used. That single step created the archive’s 26,452-strong Illustrations Library.

The Illustrations Library: 26,452 stories, analogies, metaphors and quotations, filterable by type, topic and Bible book.
26,452 illustrations mined from the sermons, each linked back to the moment it was spoken.

Ask the Archive, held back

We also built a research tool that answers a plain-language question entirely in the preacher’s own verbatim words, each quote linked to its audio. An archive of a real person’s theology can’t afford an AI that paraphrases, so the rules are strict:

  • Verbatim only. Every quoted block must be an exact substring of one real transcript paragraph.
  • Curator, not speaker. The AI may only write short bridges between quotes; a linter rejects any doctrinal wording it wasn’t given.
  • Care rails. A keyword filter routes a person in crisis to a gentle care response, even if the AI is offline.

It is built and ready to switch on.

Built to outlive us

The archive lives at the church’s own domain and depends on no single company staying alive. Audio sits on independent storage behind its own hostname. The database backs up nightly, the audio replicates weekly to a second bucket, and a full restore was rehearsed end to end: 338,688 paragraphs recovered from a real backup in about 19 minutes. It scores 100 for Accessibility, Best Practices and SEO in Lighthouse, and its word search keeps working even if the server is offline.

For engineersTranscription, enrichment, search and the player

Transcription. faster-whisper with whisper-large-v3 (float16, beam 5, VAD) across 12+ rented GPU instances, sharded by modulo with a mop-up pass. Segments were regrouped into paragraphs on a 1.5s+ pause, four sentences or 45 seconds. About $20 of compute for the corpus.

Enrichment. gemini-2.5-flash (temperature 0.1, strict JSON) returned summary, 5–12 topics from a 180-theme vocabulary, classified Scripture references, an outline with paragraph ranges, illustrations, verbatim key quotes and application points. Each carries a paragraphIndex, the hinge for click-to-play. References were checked against 66-book canonical data and quotes against the transcript. About $50.

Search. Pagefind for static full-text search; Fuse.js for faceted filtering; for Ask, 338,688 paragraphs embedded as 768-dimension vectors in PostgreSQL + pgvector (HNSW), fused with a FlexSearch keyword index by Reciprocal Rank Fusion.

Site. Astro 6 pre-renders 8,297 pages, with Preact islands only where needed. The audio player keeps its state in Nanostores so it survives navigation, and uses the Media Session API for lock-screen controls. Hetzner and Caddy, Cloudflare R2 for audio, Doppler for secrets.

We take something hard, a huge, messy, important archive, and use AI with real craft and real restraint to make it useful, beautiful and built to last.

Got an archive, a dataset, or an idea too big to sort by hand?

That’s exactly the kind of work we love. Tell us what you’re sitting on, and we’ll show you what AI, used well, can make of it.