AI invoice processing, PDF to ERP
Invoices read by an AI model against a strict schema, routed for one-click approval in Slack, and synced into the ERP with an append-only audit log. 95%+ first-pass accuracy, in daily use.
Available for remote roles · Lahore, Pakistan
I'm Amir Shahzad Hashmi, an AI engineer. I design agentic LLM systems, evaluation infrastructure and workflow automation, plus the FastAPI, Node.js and React applications around them. Currently building AI support automation for Nexar.
Selected work
Client systems I designed and built end to end. Most code is private, so each case study walks through the problem, the design and the outcome.
Invoices read by an AI model against a strict schema, routed for one-click approval in Slack, and synced into the ERP with an append-only audit log. 95%+ first-pass accuracy, in daily use.
A two-tier agent on the Anthropic API: a fast model routes, a larger model reasons only when needed, answers are grounded in real help content, and clear rules hand off to a person.
A central internal operations app connecting the store, help desk and fulfilment provider. Cancellations and refunds run from one screen with permission checks and a BigQuery audit log.
Automated refund flows for whole-invoice, partial and line-item refunds on the Recurly API. Every refund is validated against the live invoice before any money moves.
A scheduled n8n workflow that finds and enriches leads, scores and tags each one with an AI step, drops them into the CRM and notifies the team. Hands-off, every day.
RLHF benchmark tasks with reference solutions and automated graders, including multi-step agentic tasks, plus the harness that scores runs.
A dashboard that turns workforce data for 500+ employees into performance patterns and retention signals using LLM-driven analysis.
Open source
Activity and stars are read live from GitHub when the page loads.
A studio for defining AI agents as data, running them in Docker sandboxes, and letting a supervisor agent plan, review and ship work to GitHub with minimal human input.
Simulate a business decision before you make it. Five agents plan, research, run a digital twin and a Monte Carlo simulation, then critique the result; a human approves. lablab.ai × AMD hackathon.
A vision-language model served from a Docker container on AMD ROCm, reading US and Chinese licence plates and road signs from real photos. 22/22 on real photos, under 1 second per image on an MI300X.
An orchestration platform that turns a one-paragraph product description into requirements, tasks, code, tests and reviewed pull requests.
Record, inspect and replay agent runs so you can debug one bad decision without rerunning the whole thing.
Experience
From freelance WordPress to full-stack MERN work to AI systems in production.
More about meSep 2025 → Present
Coders Alley · Full-time, remote. Client: Nexar.
Jul 2021 → Aug 2025
Independent and local clients
Jun 2017 → Jun 2019
Freelancer.com.pk
Freelance
I take fixed-scope projects through PeoplePerHour and Freelancer.com: automation scripts, AI integrations, data pipelines and small web apps. Escrow-backed, with a clear brief and delivery date.
Toolbox