Amir Shahzad Hashmi

AI engineer building LLM agents, RAG pipelines and the automation around them, in production.

I design agentic LLM systems, evaluation infrastructure and workflow automation, plus the FastAPI, Node.js and React applications around them. Working professionally since 2021, currently as a full-stack automation engineer for Nexar, with open source projects on the side.

Email me Download CV GitHub LinkedIn LeetCode

Lahore, Pakistan. Open to remote roles.

Customer message embedded chat widget Haiku: classify and route Sonnet: agentic reasoning + tool calls Knowledge retrieval Storefront API FastAPI + SQLite Grounded reply to the customer

Hierarchical agentic RAG support agent, built on native Anthropic API tool calling.

Work

AI-powered customer support agent

Client: Nexar

A hierarchical agentic RAG system that answers customer support requests through an embedded chat widget.

Problem
Support requests needed accurate, grounded answers without routing every message through the most expensive model or a heavyweight orchestration framework.
What I built
A hierarchical agentic RAG pipeline on native Anthropic API tool calling, with no LangChain or third-party orchestration layer. Haiku classifies and routes incoming requests; Sonnet handles agentic reasoning and tool use. Delivered with a full-stack demo storefront (React front end, FastAPI back end, SQLite) and an Intercom-style embedded chat widget.
Stack
Python, FastAPI, SQLite, React, Anthropic API

Client work; code is private. Happy to walk through the architecture on a call.

NexClick Ops: internal support tooling platform

Client: Nexar

One internal interface for support agents to trigger back-end operations across four systems.

Problem
Support agents were handling order cancellations, refunds and fulfillment checks by hand across separate platforms.
What I built
A permissions and actions platform that gives agents a single interface and integrates Shopify (order cancellation, refunds, metafields), Intercom (attribute writes), Fulfillrite (fulfillment polling) and BigQuery (audit logging), replacing fragmented manual processes.
Stack
Shopify API, Intercom API, Fulfillrite, BigQuery

Client work; code is private. Happy to walk through the architecture on a call.

Benchmark tasks and evaluation infrastructure for LLM training

Contract work

RLHF benchmark tasks and the automated grading used to train and test large language models.

Problem
Model teams need realistic, verifiable tasks, including multi-step agentic ones, with grading that does not depend on a person reading every transcript.
What I built
Benchmark tasks with reference solutions and automated graders, including agentic tasks where the model has to use tools across several steps, and the evaluation harness that scores runs and surfaces failure modes.
Stack
Python, LLM APIs, automated grading

Task content is not public. Happy to talk through the evaluation approach.

AI-powered workforce analytics dashboard

A dashboard that turns workforce data for 500+ employees into performance patterns and retention signals.

Problem
Management needed a readable view of performance patterns and retention risk across a large workforce.
What I built
A dashboard processing data for 500+ employees, using Python and LLM-driven analysis to surface performance patterns and flag retention risk for management decisions.
Stack
Python, LLM-driven analysis

Projects

Open source work on GitHub. Activity and stars are read live from the GitHub API when the page loads.

Atelier

An orchestration platform that turns a one-paragraph product description into requirements, tasks, code, tests and reviewed pull requests. Agents work in isolated folders and every task goes through one routing function, so a supervisor can slot in later.

Stack: Python, FastAPI, OpenAI API, Alembic, Next.js.

On GitHub

AgentRewind

Record, inspect and replay agent runs so you can debug one bad decision without rerunning the whole thing. A small Python SDK wraps LLM and tool calls as spans; a web UI steps through them.

Stack: Python SDK, FastAPI, React.

On GitHub

DBPort

A self-hosted web app that connects to any database with credentials and lets you browse, query and edit it with no setup code. Introspects schemas for PostgreSQL, MySQL, SQLite, SQL Server and MongoDB.

Stack: Python, SQL and MongoDB drivers, React.

On GitHub

NexJob

A Chrome extension that fills any job-application form from your saved profile. The LLM only does semantic field matching, so it works on sites it has never seen. Profile and API key stay in the browser.

Stack: JavaScript, Chrome extension APIs, OpenAI API.

On GitHub

LeadBridge

A Chrome extension that captures structured lead data from LinkedIn, Upwork and Wellfound into a shared Google Sheet. Salespeople review and edit the extracted fields before anything is saved.

Stack: TypeScript, Chrome extension APIs, Google Sheets API.

On GitHub

RyxoGo

An experimental Go-first frontend framework that compiles to WebAssembly: signals, components, routing, server-side rendering, i18n and an MCP server so AI tools can drive it.

Stack: Go, WebAssembly, MCP.

On GitHub

Trend-following trading bot

A crypto bot on Freqtrade (Donchian breakout, 4h timeframe) deployed with Docker, plus a news-watcher service that uses LLM classification to pause trading around macro events. Running in dry-run while the strategy proves itself.

Stack: Python, Freqtrade, Docker, LLM classification.

Private repository.

.dai (contributor)

An open plain-text format for AI memory. I contributed two merged pull requests: keeping setup-written files out of git repositories, and a clearer hint after a skipped history conversion.

Stack: Node.js CLI, MCP server.

On GitHub

Earlier projects

Peace-Market.com

E-commerce platform with a multi-level referral system and JazzCash payment integration. Built the backend logic, front-end UI and server deployment.

Stack: React, Node.js, MongoDB, JazzCash API.

peace-market.com

SecureChat

Real-time web chat application on the MERN stack with WebSockets, covering live messaging, user authentication and message persistence.

Stack: React, Express, MongoDB, WebSockets.

GitHub repository

Experience

  1. Software Engineer (AI, Automation & Web Development), Coders Alley

    Sep 2025 to Present

    Full-time, remote. Client: Nexar.

    • Design and build Python back-end services and automation workflows that integrate Shopify, Intercom, Recurly and Fulfillrite through their APIs, replacing manual, cross-platform support and operations processes.
    • Develop scalable back-end and front-end components in Python (FastAPI) and JavaScript (React), including internal tooling that gives support agents a single interface for back-end operations across systems.
    • Ship AI-powered support automation and generative AI features into production, built on the Anthropic API with native tool calling rather than third-party orchestration frameworks.
    • Own features end to end, from design and API integration through production release, in a remote client team.

  2. MERN Stack Developer, independent and local clients

    Jul 2021 to Aug 2025

    • Built and deployed full-stack web applications for local clients with MongoDB, Express, React and Node.js.
    • Designed database schemas and back-end APIs, plus Firebase authentication and real-time features over WebSockets.
    • Delivered workflow automation with n8n and Zapier alongside client projects.
    • Handled server setup, deployment pipelines and basic DevOps tasks for stable, scalable production performance.

  3. Freelance WordPress Developer, Freelancer.com.pk

    Jun 2017 to Jun 2019

    • Developed and customized responsive WordPress sites for clients: theme customization, plugin integration and performance optimization.

Skills

  • Languages: Python, JavaScript, TypeScript, Go, SQL, Java, C#, C/C++
  • Back end and APIs: FastAPI, Node.js, Express, REST APIs, WebSockets, webhooks, third-party API integration (Shopify, Intercom, Recurly, Fulfillrite, Google Sheets), Firebase
  • AI and automation: Anthropic and OpenAI APIs (native tool calling), agentic RAG, LLM evaluation and automated grading, MCP servers, LLM pipelines, workflow automation, n8n, Zapier, Vertex AI
  • Front end: React, Next.js, Redux, HTML, CSS, Tailwind CSS, Bootstrap, Chrome extensions, WebAssembly from Go
  • Databases: PostgreSQL, MySQL, SQLite, SQL Server, MongoDB, Firebase, BigQuery, schema design
  • Cloud and DevOps: AWS, GCP, Docker, CI/CD with GitHub Actions, server setup and deployment pipelines
  • Practices: object-oriented design, data structures and algorithms, scalable architecture, Agile collaboration, Git and GitHub

About

Amir Shahzad Hashmi

I'm an AI engineer in Lahore, Pakistan. For the past year I've worked with Nexar through Coders Alley on Python back-end services, third-party API integrations and internal tooling, including an agentic RAG support agent that runs in production. Alongside that I write benchmark tasks and evaluation infrastructure for LLM training, and I keep a few open source projects moving: an agent orchestration platform, an agent run recorder, a database browser and a Go-to-WebAssembly framework. Before that I spent four years building MERN-stack applications for local clients, from database design to deployment.

I like owning a system from data model to production and keeping it maintainable once it's live, and I'd rather build directly on a well-understood API than hide it behind an abstraction layer I can't debug. Off the clock: video games, travel, reading and philosophy.

Education

BS Computer Science, COMSATS University Islamabad (Lahore Campus), 2021 to 2026.

Certifications

  • Supervised Machine Learning: Regression and Classification. DeepLearning.AI and Stanford Online. Issued Apr 2024. Credential ID QXZ3LNBFFF7S.
  • Introduction to AI on AMD AI PC. AMD AI Academy. Issued Sep 2026.

Languages

English (highly proficient), Urdu (native).

Contact

The fastest way to reach me is email. I reply within a day.