cd /news/ai-tools/how-forestrydiary-com-was-built-with… · home › topics › ai-tools › article
[ARTICLE · art-172111] src=orner.com ↗ pub= topic=ai-tools verified=true sentiment=↑ positive

How forestrydiary.com was built, with AI prompts

Lance Orner, lead developer of the OCCA Database Sizing Project, built forestrydiary.com by feeding four prompts to Codex, turning 18 years of U.S. Forest Service ranger Reuben P. Box's pocket diaries from 1927 to 1945 into 7,488 pages covering 217 months and 4,656 days. Codex wrote Python to scan and crop each page into a local SQLite database, Mistral transcribed the 1930s handwriting and a later pass extracted 413 people, 70 places and 50 events, and Flask templates plus flask freeze generated about 8,000 static pages and 15,000 images (840 MB) served by nginx on one small OCI compute instance. The agent configured the server over SSH, and the fully static setup kept CPU load at 0.05 under peak loads of thousands of requests a minute.

recommendation

Pre-render database-driven content to static HTML with flask freeze and rsync only changed files to a single nginx instance; the source saw peak loads of thousands of requests a minute at 0.05 CPU load.

more in playbook: Running AI in production
read3 min views2 publishedNov 18, 2026
How forestrydiary.com was built, with AI prompts
Image: Orner (auto-discovered)

forestrydiary.com

With AI prompts

Lance Orner · OCCA Database Sizing Project, Lead Developerocca.us.oracle.com

Eighteen years of a Forest Service ranger’s pocket diaries, 1927 to 1945, and the four prompts that put them online.

217 months

4,656 days

7,488 pages

Step one

Scan the paper #

Prompt to Codex

Create a Python program to capture an image from my document scanner, trim and crop, then add to a local sqlite database.

  • Eighteen years of monthly pocket diaries, one notebook per month, kept by Ranger Reuben P. Box of the Lassen National Forest.
  • Every page becomes a row : a front and back image, trimmed to the page edge, keyed by a catalogue number so it can always be traced back to its notebook.

Step two

Teach a machine to read 1930s handwriting #

Prompt to Codex

For each entry in table, use Mistral to transcribe image and put results into new column in database

  • Mistral reads each page and types out the entry, including the hours table at the bottom of each day.
  • A later pass pulls out the names : 413 people, 70 places and 50 events, with coordinates for the places.

Transcription, January 28, 1931With Asst. Hufford we drive as far as we can go down. Walk over old road to Old S.C. and thence up Big Chico Ck. to Little Smoky and back up 9 mile Ck to Rd. and then to East. Got to camp at 9 P.M. We are looking out and possibilities and burn snags along roads. Fire Prevention 10 hrs. Burn about 50 snags.

Step three

Generate the pages #

Prompt to Codex

Create Python app using Flask that resizes all images to 800px width, and create People, Places, and Events pages from database entries that have been computed. Deploy using rsync to copy static pages to compute instance.

Database

*→Flask templates→8,000 static pages→rsync→*OCI compute

  • One page per month, day, person, place and event , rendered from the database. About 8,000 pages and 15,000 images, 840 MB in all.
  • This renders as a set of static HTML/CSS pages , which is very lightweight to serve, usingflask freeze to generate the files.Hugo also works, and does for my other sites.
  • Nothing runs online but a file server. A corrected transcription means a one-minute rebuild and an rsync that carries only what changed.

Step four

Stand up the server #

Prompt to Codex

Use ssh ociserver to configure compute instance with nginx and log all traffic and referrals.

  • One small OCI compute instance running nginx, serving the static files with long cache headers on the scans and a Let’s Encrypt certificate.
  • Access logs with referrers are what tell us who reads the diaries: search engines, genealogy forums, local history groups, and which bots to turn away.
  • Static sites are easy to serve. Peak loads of thousands of requests a minute only push the CPU load to 0.05.
  • The agent did the setup over SSH : packages, vhost, certificate, log rotation. Nothing was typed on the server by hand.
── more in #ai-tools 4 stories · sorted by recency
── more on @lance orner 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/how-forestrydiary-co…] indexed:0 read:3min 2026-11-18 · —