Skip to content
▓ about_me.txt

Hasnain Mubashir

Solutions Engineer

I take on whatever the problem is and own it end to end. Web apps and UI in Next.js, React and TypeScript. Type-safe backends and APIs. Training and evaluating LLMs. Squeezing milliseconds out of Core Web Vitals. No handoffs and no excuses — if it needs building or fixing, it’s on me. Based in Lahore, Pakistan.

Most of my work sits where those things meet. An AI feature is a front end, an API, a data store and an evaluation loop, and it tends to fail at the seams between them rather than inside any one of them. Working across the whole thing means the model evaluation is informed by what the product actually does, and the interface is built by someone who knows what the model gets wrong.

Practically, that means one person to brief, one person accountable, and no gap between “the API team said” and “the front-end team said”. It suits teams who need a thing built properly and do not want to run an agency to get it.

LOCATION:
Lahore, Pakistan
EMAIL:
talktome@hasnainmubashir.com
▓ skills.idx

Skills and technologies

Next.jsReactTypeScriptNode.jsPythonFastAPITailwind CSSPostgreSQLRedisGraphQLREST APIsLLM EvaluationPrompt EngineeringRubric ScoringRed-TeamingSanity CMSDockerVercelGit
▓ hasnain.bmp
Hasnain Mubashir

▓ system_stats.log

free tools
34
reply time
<24h
uploads
0
▓ ai_llm.log

AI & LLM work

Alongside the engineering work I train and evaluate large language models — a full year of it, full time.

  • Training and evaluating voice-assistant models on intent understanding, entity extraction and tool/function-call correctness.
  • Reviewing multi-turn conversations for context retention, disambiguation and recovery from misheard or incomplete requests.
  • Rubric-based scoring and comparative ranking of candidate model outputs, held consistent against project quality standards.
  • Red-teaming and edge-case hunting — ambiguous phrasing, transcription noise, out-of-scope requests, unsafe or privacy-sensitive queries.
  • Coding and reasoning task review for benchmark work: rubric alignment, hidden failure modes, false passes and false fails.
  • Leading a review team — calibrating reviewers, setting acceptance criteria and keeping annotation decisions aligned.
▓ proof.log — go and check

Don’t take my word for it

Anyone can list technologies. These are things you can open right now and judge for yourself.

  • This site

    Server-rendered Next.js with no animation library, per-route code splitting, self-hosted fonts and performance budgets. The tool pages ship only their own JavaScript. You are looking at the work sample.

    Read the build notes →

  • 34 browser tools

    Image conversion with batch processing and in-browser background removal, generators, and calculators — all running client-side with nothing uploaded. Built, shipped and maintained, not prototyped.

    Open the toolbox →

  • Open source

    Code, commit history and the things that did not work are all public. The fastest way to judge whether someone can build is to read what they have already built.

    GitHub profile

What I do

📁 ui_dev/

Websites & UI Development

Responsive, accessible UIs built in Next.js with component-driven architecture and CSS-driven motion.
📁 ai_llm/

LLM Training & Evaluation

Rubric-based scoring, prompt and response evaluation, tool/function-call correctness, red-teaming and edge-case detection for model training pipelines.
📁 api/

Backend & APIs

REST/GraphQL services with Node.js, Postgres, Redis, auth, rate-limits, and observability.
📁 perf/

Performance Audits

Core Web Vitals optimization, Lighthouse budgets, RUM dashboards, and CI regression guards.
📬 lets_talk.msg

Want to work together?

A new product, a rescue on something half-finished, a performance audit, or an LLM evaluation pipeline — if it needs doing, tell me about it.

▶ Get in touch