Skip to content
Muhammad Safyan

Videos

Watch me walk through the work

Real projects, recorded on screen: what was built, the hard part, and how I solved it. Jump to any chapter.

Web apps, back ends, data and automation — a 3-minute intro

3:18 min

A quick tour of the kind of work I do: a car-listings platform, turning spreadsheets into an inventory app, a Swedish services marketplace, AI matching and upgrading a live platform.

  1. 0:14Car-listings platform
  2. 0:43Inventory: spreadsheets to an app
  3. 1:15Swedish services marketplace
  4. 1:43AI user matching
  5. 2:21Upgrading a live platform
  6. 2:49Working together

Full-stack work: six products and the hard part of each

15:08 min

Six production apps and the hardest problem in each: API limits and PgBouncer, Stripe webhooks, stats computed at upload, Pinecone matching, two-way WooCommerce sync and a Laravel takeover.

  1. 0:19CarLister — deal finder, API limits, PgBouncer
  2. 2:46Helpicon — Stripe webhooks, never trust the front-end price
  3. 4:43CVR — Angular, stats computed at upload
  4. 6:24Abundance — Pinecone matching, bulk save 20 s to 2 s
  5. 9:13BrickDistrict — spreadsheets to app, two-way WordPress sync
  6. 10:48Laravel + React (Inertia) takeover, missing migrations
  7. 13:42Why I start with a meeting

AI features in production: LLM moderation, Pinecone matching, RAG and cost

15:10 min

AI features running in real products: LLM content moderation, auto-written descriptions, people matching with Pinecone, a RAG chatbot kept away from the database, and cutting AI cost.

  1. 0:13Marketplace moderation with an LLM
  2. 1:56Why the first prompt blocked real posts
  3. 3:02Auto-writing task descriptions
  4. 3:29People matching with Pinecone
  5. 4:01How matching works
  6. 6:14Match threshold, namespaces and weights
  7. 7:41Chatbot: help docs or your own data
  8. 8:11Keeping an AI agent away from your database
  9. 10:11Cutting AI cost: the right model for the job
  10. 12:56Computer vision: agility drills and rugby

Making manual work disappear: from spreadsheets to a system that runs itself

7:37 min

Replacing spreadsheets and manual reports with automation: an inventory app with two-way WooCommerce sync for BrickDistrict, and Excel match reports turned into automatic ones.

  1. 0:14BrickDistrict: spreadsheets and Sortly to one app
  2. 1:34Two-way sync with the WooCommerce shop
  3. 2:06One source of truth for stock
  4. 2:51A 30-second save: the plugin behind it
  5. 3:35Geo-blocked APIs
  6. 4:04More than inventory
  7. 4:55CVR: Excel reports to automatic reports
  8. 6:21Data warehouses behind the automations

Taking over an existing app: what I check in the first week

15:00 min

What I check when I inherit a codebase: reproducing bugs first, auth gaps, N+1 queries, tenant isolation, private files, logs, silent cron failures and shared databases — plus a real takeover.

  1. 0:15Step 0: understand the business
  2. 1:01Reproduce before touching
  3. 1:48Report too big: PNG vs JPEG
  4. 2:55Auth: suspended but still logged in
  5. 4:19N+1 queries
  6. 5:31Tenants: RLS and connection pools
  7. 8:28Sensitive data: file names, public vs private
  8. 9:48Logs: keep them, keep tokens out
  9. 10:49Failing every hour
  10. 11:47Two environments, one database
  11. 12:30A real takeover: database to RDS, files, migrations

Have a project in mind?

Tell me what you are building and where it hurts, and I will come back with questions or a plan.