I'm a full-stack engineer with over 14 years of experience. I work in TypeScript and Python, and most of my recent work involves LLMs.

I'm open to full-time roles and contract work. I work remotely from Pakistan, UTC+5.

• React and Next.js frontends, and Node, NestJS and FastAPI backends on Postgres.
• LLM features, including prompts, evals, model benchmarks, retrieval over large document sets, human review for low-confidence output, and pipelines that send most calls to a cheaper model.
• Performance problems such as N+1 queries, missing indexes, large bundles and extra React re-renders.
• Deployment and monitoring with Docker, GitHub Actions, AWS, Vercel and Sentry.

I've built products for companies in HR, healthcare, procurement, hospitality, food service, retail, auto parts and consumer AI.

Before 2021 I built Vue, Nuxt and Laravel apps, including ERPs, point-of-sale systems, auto-parts stores and booking sites.

• I raised LLM extraction accuracy from 63% to 92% against hand-checked data by benchmarking models and redesigning the pipeline. Cost per document went down at the same time.
• I built an LLM service that replaced a retired third-party service inside a client's core system. It extracts data against a versioned schema, runs OCR on scanned files, checks the output, and returns results by signed callback.
• I designed the architecture for a platform that turns inspection reports into supplier analytics. I wrote its C4 diagrams and six architecture decision records. I then compared three extraction pipelines: rule-based NLP, AWS Textract with Comprehend, and an LLM with a verification step and a human review queue. Later I built its reporting system, which renders dashboard charts on the server with Playwright, exports reports automatically, and recovers stuck jobs.
• I've worked with one catering company since 2016. I started with their staff-hours app and now maintain a monorepo of eight apps for inventory, recipes, fleet, packing and invoicing. I wrote 96% of its 4,000 commits.
• I set up bundle analysis on a large Next.js and MUI app, moved its heavy modules into lazy-loaded chunks, and cut extra re-renders with memoization and zustand selectors.
• I've used AI coding tools since the GitHub Copilot technical preview in 2022. I now work mostly through Claude Code, Codex and Cursor, and I write the agent instructions, skills, coverage gates and eval loops my teams use with them.

TypeScript, Python, React, Next.js, TanStack, Vue, Nuxt, Node, NestJS, FastAPI, Laravel, Postgres, Redis, Prisma, tRPC, OpenAI, Anthropic, OpenRouter, LlamaIndex, Docker, GitHub Actions, AWS, Vercel, Sentry, OpenTelemetry, Grafana.

Email me at salalaslam@gmail.com.