Full-stack developer with an MCA and 3+ years of dedicated Python development. As a Handshake AI Fellow, I design software tasks, reference solutions, and automated tests that train and evaluate frontier AI coding models. I build in Python, .NET, and SQL, and I write about it.
I’m a software developer and AI-evaluation specialist with a Master of Computer Applications and 15+ years across software engineering, database work, data quality, and research. I’ve built full-stack web applications and backend systems in Python, .NET, C#, and Java, and tuned complex SQL and Oracle PL/SQL workloads for performance.
Today I’m a Handshake AI Fellow, creating challenging software tasks that train and evaluate frontier AI coding models — designing the problem, setting up a reproducible environment, writing the reference solution, and building tests that accept every valid solution while rejecting subtly broken ones. Before that, I evaluated multilingual LLM and speech output at Appen. It sits right where software engineering meets AI quality, and it’s the work I find most interesting.
A real-time railway-signaling game for 2–4 players where each player’s working orders appear only on a neighbor’s screen, so they have to talk to get the trains through. A server-authoritative Durable Objects backend sends each seat only its own view; a seeded generator solves every shift before printing it (verified on 600 generated layouts), and a capacity fix raised simulated completion from 2/400 to 398/400 nights.
A Slack app whose /wordcount slash command posts a draft’s word count and how far it is over or under a limit. Built for least privilege: it replies in the command’s own response, so it needs only the commands scope and stores no bot token, and it verifies every request with Slack’s signing secret (HMAC-SHA256, 5-minute replay window).
A dependency-free harness for scoring LLM answers against a gold set, with per-question match rules (exact / contains / numeric), gold-file validation, actionable failure output, and a CI-friendly exit code.
A compact task-tracker Web API in a single file — full CRUD with validation and correct status codes — plus integration tests that boot the app in memory and exercise the real endpoints.
Runnable examples that turn a slow row-by-row load into fast batched processing with BULK COLLECT and FORALL, plus SAVE EXCEPTIONS error handling — demonstrated on a 200k-row table.
A full-stack web portal, built during earlier software-development work, for submitting municipal complaints and tracking their resolution — with an admin dashboard that routes each complaint to the right department and monitors real-time status. Full-stack build across backend, database, and UI.
More at dev.to/zahid23saim
Languages: English, Hindi, Tamil, Marathi — professional fluency (native / near-native).
Open to remote, project-based software and AI-evaluation work.