Skip to main
Omar Nagy.

The notebook · production AI · real numbers, no hype

Fieldnotes.

Engineering notes on building and operating products, including AI. Read the decisions, tradeoffs and measurements behind the case files.

14 entries on file · engineering decisions + evidence

The register · newest first

01Receipts~6 minfiled 2026-09-291A feedback button is only useful when it carries the right contextA report that says “it broke” usually creates another round of questions. I traced feedback across several products I have built and found a reusable pattern, copied adaptations, separate implementations, and places where error logs are being mistaken for user feedback.02Receipts~7 minfiled 2026-09-29RTLArabic-first software is a workflow decision, not a translation taskA right-to-left interface can still feel like a left-to-right product wearing Arabic labels. The hard work is following how people move through a real task, then making the language, controls, roles, and handoffs fit that routine.03Receipts~6 minfiled 2026-09-29180A study session should not start with an incomplete examA learner asked for a full exam. One production response contained only part of the expected question set, yet the start action was still available. The fix was a clear readiness gate and retry path, followed by a synthetic check of both incomplete and complete responses.04Positioning~7 minfiled 2026-09-294Before adding AI to support, make the fix verifiableWhen a user reports a problem, AI can help organize the evidence. But if the report lacks context, the owner is unclear, and nobody checks the repair, automation only makes a weak process move faster. Here is the repair loop I design first.05Receipts~9 minfiled 2026-06-10100% → 17%llm-eval-ci: the gate that fails the PR when your LLM quietly regressesEvery part of an LLM product is gated except the part that matters. llm-eval-ci is the open-source CI quality gate I shipped today: a golden set scored by six graders, exit 1 when answer quality drops. In the demo, a harmless-looking prompt rewrite falls from 100% to 17% and the build goes red.06Receipts~10 minfiled 2026-06-107Mnemonic: self-hosted, categorized memory for AI agentsA small self-hosted FastAPI memory server for AI agents, built on mem0 and Qdrant. It auto-sorts every memory into seven semantic categories, serves a tiered L0/L1/L2 context tree instead of dumping the whole pile, and keeps the conversation history on your own box. About $2/month self-hosted versus $20+/month for the cloud memory APIs.07Receipts~3 minfiled 2026-06-10AdoptionA family-shop ordering system and the limits of replacementHistorical family-business engineering: an Arabic-first custom ordering system, separate from RetailOS.08Receipts~10 minfiled 2026-06-1037%Maintaining a 10,992-question medical exam bank (and rejecting 37% of the fixes)In the June 10, 2026 snapshot, MedPrüf held 10,992 active German-language medical exam questions, maintained from Egypt by a non-doctor. The quality lives in the correction pipeline: 91 suggested fixes triaged, 37% rejected on review, 67 human-reviewed AI explanations, and zero runtime AI calls served.09Receipts~3 minfiled 2026-04-30ValidationSupplier-data extraction: calibration, validation and reviewHistorical methods from paused Bridge Sourcing: labelled calibration data, deterministic validation and human review.10Positioning~3 minfiled 2026-04-30CairoWorking across time zones from EgyptRemote product engineering from Egypt: agreed availability, clear project stages and usable handoffs.11Receipts~3 minfiled 2026-04-30RoutingModel-cost controls: routing, caching and context budgetsHistorical engineering methods for controlling model spend, with a tested illustrative router. Quantified outcome claims retired pending measurement receipts.12Positioning~9 minfiled 2026-04-303Fractional CTO vs agency vs full-time AI hire: buyer-side mathIf your board just asked about AI strategy and you have 5–50 people, the highest-EV next move is a fractional senior, assuming you do diligence. The math on what each of the three paths costs you when it doesn't work.13Receipts~13 minfiled 2026-04-3014dHarmonia: a single-tenant luxury salon system, end-to-end in 14 daysSingle-tenant on purpose, against every SaaS instinct. An agency would have quoted 8–12 weeks at around $40K for the same POS + booking + payroll + receipts + RTL + PWA scope. I shipped v1 solo in 14 days, then a v2 sweep two days later that added role-aware UIs, finances, audit, and commission settlement. The architectural call is what made the calendar fit.14Positioning~8 minfiled 2026-04-305Production AI in 2026: five shifts most teams are still ignoringMost teams shipping AI in 2026 are still solving 2024 problems with 2024 tools. Here are the five shifts that flipped (prompt caching, agents, evals, tool-calling, observability) and what each one costs to ignore.

Newsletter · hand-run, no automation

New entries in your inbox if you ask for them.

Each entry walks one production AI move: what it contributed, what it broke. Drop your email for the early issues; otherwise it lives here.

Your email is used to request the manually managed newsletter. When configured, the request is forwarded to Omar through Telegram. This does not automatically subscribe you. Privacy and contact.