Umer Arshad

I make software work where it is hardest to ship.

On a microcontroller. Inside a live market data feed. Across a vehicle network. Eight years of production C and C++, and now I bring AI into those same places.

AI that fits inside real engineering constraints

I use AI coding assistants every day on systems code, and I choose where each model runs. Proprietary FinTech and firmware code stays on local models. Architecture and review questions go to frontier models.

What I work with

  • AI-assisted development
  • Local LLMs with Ollama
  • ONNX Runtime in C++
  • Edge inference
  • Agent tooling

Pick another layer to see what I have built there.

Where I fit

Three kinds of work I am looking for. Each one draws on the same habit: going to where the system really runs and making it work there.

Forward deployed engineering

I have shipped into trading infrastructure, vehicles and consumer devices, each with its own rules, legacy and risk. I am comfortable sitting with a customer's engineers, learning their system fast and putting AI into it without breaking what already works.

AI-assisted development

I build with AI assistants daily in C, C++ and Python. I know where they speed a team up, where they produce confident mistakes in systems code, and how to set them up so private code stays private.

Systems, embedded and IoT projects

Freelance builds across the stack: ESP32 firmware, low-latency C++ services, IoT backends and Flutter apps. I can take one layer or own the whole path from device to cloud.

Projects

Most of this code is proprietary, so these are summaries. The bars beside each project show which layers of the stack it touched.

Real-time market data pipeline

Logiciel Services, FinTech

A production pipeline that takes exchange data in over FIX and distributes Level 1, Level 2, options and derivatives data to downstream consumers through ZeroMQ, with Redis for fast shared state. I owned it as technical lead.

  • C++
  • FIX
  • ZeroMQ
  • Redis

Technical signal engine

Logiciel Services, FinTech

A C++ engine that turns live market data into trading signals: MACD, Stochastic, Parabolic SAR and candlestick pattern recognition. I am extending the design toward model inference with ONNX Runtime.

  • C++
  • Technical indicators
  • Pattern recognition
  • ONNX Runtime

Patented high-availability cluster

Afiniti, enterprise software

A cluster system in C++ that keeps services running through node failure. The design was granted a patent. Communication is secured with TLS and the services ship as Docker microservices.

  • C++
  • Clustering
  • TLS
  • Docker

Connected lighting platform

Watts Lights, IoT

The full path for a connected lighting product: ESP32 LED controllers in C, MQTT messaging through EMQX and AWS IoT Core, DynamoDB storage, and a Flutter app for control.

  • ESP32
  • MQTT
  • AWS IoT Core
  • DynamoDB
  • Flutter

Vehicle network software

Siemens Digital Industries, automotive

Embedded software for electronic control units: CAN and LIN communication, ISO-standard bootloaders and low-level drivers.

  • C
  • CAN
  • LIN
  • Bootloaders
  • ECU drivers

How I work with AI

I come to AI from the systems side. That shapes what I care about: latency, privacy, and whether the thing still works at three in the morning.

Private code stays on private hardware

Trading logic and device firmware are a company's core IP. I run local models through Ollama for that code, and keep frontier models for architecture decisions, security analysis and long-context review.

Assistants in the daily loop

I use AI assistants for firmware debugging, C++ boilerplate, test scaffolding and code review, with editor integration. I treat the output like a junior engineer's pull request: useful, fast, and always reviewed.

Inference where the data already is

Calling an API is the easy part. I am interested in running models inside a C++ process on the hot path, and on the device at the edge, where memory and timing budgets are tight.

Experience

Four companies and three industries. The common thread is C and C++ close to the hardware or close to the clock.

Logiciel Services

Technical Lead, FinTech

Led the team behind real-time market data pipelines covering Level 1 and Level 2 quotes, options and derivatives, and maintained a proprietary C++ signal engine, on a stack of FIX, Redis and ZeroMQ.

Watts Lights

IoT Developer, connected lighting

Developed the full IoT stack for connected lighting: ESP32 firmware in C, cloud services on AWS IoT Core, EMQX, DynamoDB and Azure, Python tooling, and the Flutter mobile app.

Afiniti

C++ systems, enterprise software

Built a high-availability cluster system in C++ that was granted a patent, secured service communication with TLS and shipped Docker-based microservices.

Siemens Digital Industries

Embedded software, automotive

Developed automotive embedded software for CAN and LIN vehicle networks, including ISO-standard bootloaders and ECU drivers.

Education: BS Electrical Engineering, University of Engineering and Technology (UET), Lahore.

Have a system that needs AI inside it?

Or firmware, a C++ service or an IoT product to build. Tell me what you are working on and I will reply with how I would approach it.

umerm64@gmail.com

Lahore, Pakistan (UTC+5). Working remotely with teams worldwide.