Omar G. Nagy
AI Product Engineer
Egypt (remote, GMT+2) · omar@neurascale.org · +20 150 047 7334
omargnagy.com · github.com/omarnagy91 · linkedin.com/in/omargnagy
Summary
AI Product Engineer with 9 years shipping production software, the last 3+ building LLM systems that run in production. Founder-operator of NeuraScale: products in production and 2 open-source tools, built and operated solo, including a multi-tenant POS SaaS and a medical exam-prep platform with 10,700+ questions. Cut LLM costs 28% with measured, published methods. Full EU timezone overlap, 4+ hours with US East.
Skills
- AI / LLM
- Claude API (Anthropic), OpenAI API, multi-agent orchestration, agent tool use / function calling, RAG, vector search (pgvector, Qdrant), prompt engineering, prompt caching, eval harnesses, LLM cost optimization, Model Context Protocol (MCP)
- Languages
- TypeScript, JavaScript, Python, SQL, Bash
- Web
- Next.js (App Router, RSC, Server Actions), React, Node.js, FastAPI, REST APIs, Tailwind CSS
- Data / Infra
- PostgreSQL (RLS, migrations, query tuning), Supabase, Vercel, GitHub Actions, CI/CD, Sentry, GA4
- Testing
- Playwright, Cypress, end-to-end coverage on critical paths
Experience
NeuraScale · Founder & AI Product Engineer
- Built and operate production products solo (RetailOS multi-tenant POS SaaS, MedPrüf exam prep, Bridge Sourcing B2B, Harmonia POS, Blink cafe OS) on TypeScript, Next.js, Supabase/PostgreSQL, and Vercel, serving live paying users.
- Shipped MedPrüf, an Austrian medical-licensing exam prep platform, growing it to 10,700+ active questions across 3 exam types and 16 specialties, by building an AI content pipeline with strict-format explanation generation in German plus 7 locales, subscription payments, and a 7-tab admin dashboard.
- Cut portfolio-wide LLM spend 28% with no quality regression by building model routing, prompt caching, and context-budget enforcement; published the per-change breakdown publicly.
- Took RetailOS from hand-managed to invite-only multi-tenant SaaS in a 48-hour sprint by shipping 12 PRs and 12 PostgreSQL migrations (RLS hardening, webhook signature auth, idempotency, server-side re-pricing) with zero data loss.
- Raised extraction accuracy from 82% to 96% on Bridge Sourcing’s supplier-data pipeline (20 structured fields per record) using hand-labeled calibration sets, deterministic validation outside the model, and human-review acceptance gates; published the full engineering breakdown.
- Scaled a one-person engineering org by designing a multi-agent build harness (research, build, review, and test lanes with human approval gates); open-sourced the agent memory layer (Mnemonic) and the eval tooling (llm-eval-ci).
- Selected as a paid evaluator of agentic coding assistants (primarily Claude Code), reviewing and validating AI-generated changes across advanced production codebases (under NDA).
Independent (Consulting) · AI & Automation Engineer
- Reduced an enterprise client’s LLM spend by building cost guardrails (daily caps, cheap-prompt-first routing, context budgets) with no measured quality regression; the pattern became standard across all my later products.
- Built production web-data extraction pipelines (retry, deduplication, fallback chains) and lead-scoring microservices for SME clients pivoting into AI integration.
- Shipped a production multi-agent pipeline in early 2023 (RAG retrieval plus tool-calling orchestration with hand-rolled JSON-mode prompting and retries) before mainstream agent frameworks existed.
Independent · E-Commerce Engineering Consultant
- Drove ~35% conversion lift for seller cohorts on Taager, a $6.4M-funded social-commerce platform scaling past 100+ active sellers, through funnel instrumentation and seller-side conversion work.
- Replaced spreadsheet stock counts with a live inventory-sync engine across 4 warehouses for a textile manufacturer, including a QR-code loyalty MVP, delivered in 6 weeks.
- Migrated 3 Shopify storefronts to custom Next.js with Stripe checkout, lifting Lighthouse scores from 38 to 92.
mBalloon · Co-Founder
- Onboarded 15+ MENA SMBs onto Shopify with paid acquisition during the COVID e-commerce boom; 3 niche stores sustained 2.5-3.2x ROAS for 6+ months.
- Cut per-client setup time from 3 weeks to 5 days by building a reusable Shopify theme and funnel template.
Freelance · Front-End Developer
- Migrated a 60K-LOC jQuery codebase to React with Cypress E2E coverage on critical paths, cutting time-to-interactive from 3.5s to 2.1s with zero production breakage (additive migration, feature flags).
LibreFit · Founder
- Built a smartphone computer-vision body-measurement product end to end (Xamarin client, OpenCV pipeline, server-side calibration) reaching ±1-2 cm accuracy; sold 2 contracts to MENA tailoring shops.
Open Source & Writing
- Mnemonic: Self-hosted agent-memory layer (Qdrant vector storage, retrieval API). github.com/omarnagy91/mnemonic
- llm-eval-ci: CI-integrated eval harness for LLM outputs. github.com/omarnagy91/llm-eval-ci
- Writing with real production numbers at omargnagy.com/writing, including:
Education
BSc, Computers & Informatics · Suez Canal University, Ismailia · 2014 – 2018 · First Class Honours
Languages
English (fluent) · Arabic (native) · German (A2, learning)