Ahmad Afnan Amjad

Ahmad Afnan Amjad

Applied AI/ML Engineer specializing in production-grade agents and high-precision computer vision.

Based in: Lahore, PKMost recently: Applied AI/ML Engineer, MINDSTORM STUDIOSExperience: 4 years

Hire me

Ahmad is an experienced engineer with a strong background in deploying scalable AI pipelines and IoT platforms across diverse industries. He has a proven track record of leading technical teams to deliver high-impact solutions, from million-user support bots to safety-critical industrial firmware.

Experience

Applied AI/ML Engineer

MINDSTORM STUDIOSLahore, PKworkDec 2025 – Oct 2026

Summary

Designed and deployed AI-powered internal tooling forlevel design, marketing asset generation, and business intelligence. Engineered production-grade AI agents with custom-tuned harnesses, consistently maintaining a sub-1% failure rate at scale. Trained a team of 3 junior engineers in AI and harness engineering, enabling them to build a user-support bot serving million-user-scale applications. Subsequently led the same team to design and build an AI-assisted level generation pipeline.

What I did

  • Developed an agent harness using a cyclical benchmarking process to identify failure patterns and implement iterative fixes.
  • Implemented agent logic using LLM API calls with specific parameters, prompts, tools, and retry logic for transient failures.
  • Implemented response schemas using Gemini API's native response schema field to achieve constrained output.

Results

  • Consistently maintaining a sub-1% failure rate at scale for AI agents.
  • Enabled team to build a user-support bot serving million-user-scale applications.
AIMLHarness EngineeringAI AgentsLLM APIBenchmarkingStructured DecodingGemini API

Software Engineer (AI/ML, Computer Vision & Embedded Systems)

TEKTRACKING, LLCLahore, PKworkSep 2023 – Dec 2025

Summary

Led and shipped Crossing Vue 360 - an IoT platform forreal-time monitoring and fault-reporting ofrailway infrastructure. Built scalable Computer Vision and Edge Data pipelines delivering real-time analytics on resource-constrained devices. Trained and deployed multi-task YOLO CNN models for object and pose detection at pixel-level precision, enabling accurate automated fault reports. Independently designed and shipped firmware fortwo production systems: a Smart Kitchen Appliance (C++/FreeRTOS on ESP32) and an industrialrailway automation controller(CODESYS/Structured Text). Architected a modular ESP32 firmware design that absorbed major feature additions and a full UIredesign with minimalrework. Implemented fail-safe, safety-critical logic forthe railway controller, plus a Python/Tkinter diagnostic tool that cut testing time and errorrates.

What I did

  • Developed a first-generation automatic monitoring solution to replace weekly manual inspections by human technicians.
  • Provided real-time visibility into entire operational areas, including highly remote locations.
  • Optimized YOLO CNNs to run over ONNX runtime on constrained edge devices for near-real-time video processing.
  • Processed RTSP video streams from multiple cameras on small form factor PCs with low-end GPUs.
  • Trained and fine-tuned models to ensure consolidated reports were uploaded within minutes of an event.

Results

  • Successfully delivered the product within the allowed performance budget by fine-tuning a smaller model variant using insights from a larger model.
  • Shipped Crossing Vue 360 IoT platform.
  • Delivered real-time analytics on resource-constrained devices.
  • Achieved pixel-level precision for object and pose detection.
  • Cut testing time and error rates with a custom Python/Tkinter diagnostic tool.
Computer VisionEdge Data pipelinesYOLO CNNC++FreeRTOSESP32CODESYSStructured TextPythonTkinterONNX RuntimeRTSPQuantization

Application Developer

IBMLahore, PKworkAug 2022 – Sep 2023

What I did

  • Worked on-site at the client's office with IBM and MCB teams to resolve potential migration problems ahead of time.

Results

  • Ensured a smooth migration process with zero downtime for an ESB upgrade project with MCB Bank.
  • Proactively automated repetitive parts of the development process.
  • Increased development speed by approximately 30% through automation.

Skills

Technical

YOLO CNN
Embedded Systems
Harness Engineering
PythonPython
C++C++
LinuxLinux
Prompt Engineering
LangChain
Computer Vision
System Architecture and Design
Generative Models/LLMs
NumPy
PyTorchPyTorch
OpenCVOpenCV
Systems Engineering
Networked Applications
CC
SQLSQL
ONNX Runtime
FFmpeg
FastAPIFastAPI
PostgreSQLPostgreSQL
MySQLMySQL
FreeRTOS
CODESYS
Video Processing
Vector Embeddings
Distributed Systems
Concurrency & Parallel Processing
Algorithms and Optimisation
Shell
LaravelLaravel
LATEX
C#C#
JavascriptJavascript
PHPPHP
Structured Decoding
Ollama
Quantization
Benchmarking
LLM API

Projects

SPLIT AUTOENCODER FOR VIDEO UPSCALING

Team Leadacademic
Led a 3-person team to design and train a novel neural network architecture that incorporates a compression codec for backwards-compatible scalable video encoding. Achieved superior visualresults compared to conventional upscaling methods (e.g., bicubic interpolation) applied after decompression.
PyTorchNumPyFFmpeg

PEER-TO-PEER CONTENT DELIVERY NETWORK

Architect and Implementeracademic
Architected and implemented a decentralized P2P CDN to reduce network traffic on central servers. Leveraged Mininet to emulate a realistic network topology of up to 30 nodes; experimentally verified that the design reduced core network traffic by offloading requests to the peer network.
MininetPythonLinux

Education

Undergraduate Studies

Syed Babar Ali School of Science and EngineeringScience and Engineering

BS in Computer Science

LUMS UNIVERSITYComputer ScienceJun 2022

Coursework

Internet ResearchDistributed SystemsNetwork SecurityInternet of ThingsImage and Video CodingAdvanced ProgrammingOperating Systems
Hire AhmadReach out about a role, a contract or a conversation.For recruiters

Ahmad's twin is AI, it can make mistakes.

More people on Kanvis

Browse more peopleCreate your own Kanvis