Hi, I’m Ajmal.

Senior AI Engineer

I build LLM agents, retrieval and evals, backed by 6 years of shipping production software.

See my workDownload résumé

About

AI systems, built like software.

I’m an AI engineer at AKstudios, a technology consultancy. I build generative AI systems for clients: agentic workflows on Gemini and Claude, semantic search with hybrid retrieval, LLM document extraction, and the evals that track precision, recall and cost.

Before that I spent four and a half years on the founding team at Penny Software, shipping 500+ features on an enterprise procurement platform. I work directly with clients and own delivery from requirements to production code, infrastructure and tests.

Education

  • Master of Computer ScienceUniversity of Illinois Urbana-Champaign, Expected Dec 2026
  • BS Software EngineeringKing Fahd University of Petroleum & Minerals, 2020

Toolbox

  • Python
  • TypeScript
  • Claude
  • Gemini on Vertex AI
  • RAG and hybrid search
  • Evals
  • NestJS
  • PostgreSQL
  • AWS
  • Google Cloud
  • Docker

Experience

Six years of shipping.

  1. AKstudios

    AI Engineer · Carol Stream, IL

    • Built an LLM document extraction pipeline that pulls structured fields from unstructured input for reports and timesheets. Deterministic code owns totals and dates, so the model never does arithmetic.
    • Took an EV charging platform from zero to production as the sole engineer: a NestJS and PostgreSQL API with RabbitMQ events, a React Native kiosk with Stripe Terminal payments, 165 test files and 8 CI/CD pipelines.
    • Built Python gateway software on AWS IoT Core for a smart radiator startup’s pilots at 3 institutions, then traced 85% of missing telemetry to gateway power loss across 1,398 gateway-days.

  2. Penny Software

    Senior Software Engineer, founding team · Riyadh, KSA

    • Shipped 500+ features over 4.5 years on an enterprise procurement platform, including the marketplace used across 30+ enterprise deployments.
    • Built an unsupervised clustering engine with UMAP and HDBSCAN that groups purchase requests for the same vendor into one quotation request, aimed at unlocking bulk pricing.

  3. King Fahd University of Petroleum & Minerals

    Research Assistant · Dhahran, KSA

    Built an ETL pipeline that feeds 3 data sources into agent-based COVID-19 outbreak simulations.

  4. AmeriCloud Solutions

    Software Engineer · Schaumburg, IL

    Cut cell-site database generation for Nokia RF engineers from 15 minutes to seconds with a Flask and Pandas service.

Featured projects

Open source.

swatter-bot.

Slack bug reports in, deduplicated GitHub issues out

A Slack bot that files bug reports as GitHub issues and tells the reporter when they close.

Duplicate detection retrieves, then judges: SQLite FTS5 keyword search and bge-small embeddings merged by reciprocal rank fusion, then an LLM judge on the top 3. Model output is validated with Pydantic and retried on failure.

  • Python
  • Slack Bolt
  • Hybrid retrieval
  • LLM-as-judge
  • Pydantic
View on GitHub

flux.

Measuring whether agent ceremony pays for itself

A Claude Code plugin that guides coding agents through plan, audit and apply skills, with a benchmark harness that grades workflows on held-out tests while the model stays fixed.

Across 3 runs, the full 4-session lifecycle cost 2.8x to 2.9x a single session for identical acceptance scores. Planning is now opt-in.

  • Python
  • Claude Code
  • Evals
  • Benchmarking
View on GitHub

codometer.

A flight recorder for coding agents

Reads Claude Code, Codex CLI and opencode transcripts to track token cost per session and map which files agents change, on a live 2.5D map.

Local-first and deterministic, with zero network egress.

  • TypeScript
  • Agent observability
  • Token cost
View on GitHub

Contact

Let’s talk.

Hiring for AI engineering, or building something with LLMs? Email is the fastest way to reach me.

GitHubLinkedInRésumé (PDF)