UM usman.malik
ESC
↑↓ navigate ↵ select ⌘K or Ctrl+K anytime

Open to Lead, Manager & Senior roles · Remote

Usman Malik. I build AI that pays for itself.

> AI Engineering Lead · Agentic AI

I find where AI can save a business money — then research it, build it, and ship it to production. Nearly 8 years taking AI from research to real-world deployments, from real-time computer vision on edge devices to agentic LLM systems, and leading the teams that build them.

See my work Résumé

  • 8 yrs

    Applied AI

    From data pipelines to production LLM systems

  • 400 +

    Models shipped

    Trained, optimised and deployed to production

  • 50– 60 %

    LLM cost cut

    By matching each use case to the best-value model

  • 4 ×

    CV throughput

    Per camera, on the same hardware

01About

From pixels to agents.

I’m an AI engineer and team lead who helps companies figure out where AI can save them money, builds it — and fixes it when it isn’t working.

My career started in the data trenches: in 2018 I built and led an annotation team at DARVIS, then moved into computer vision, training and deploying 400+ production models for real-time detection, tracking and pose estimation — on GPU servers and on Jetson edge devices.

As LLMs matured, I followed the frontier. Today I lead LLM R&D at Enghouse Interactive: benchmarking closed and open-source models to match each use case to the best-value option (cutting costs 50–60%), and building agentic assistants and RAG pipelines for customer service teams. Along the way I raised video analytics throughput 4× per camera at DP World and led AI engineering at a YC S25 startup.

I sit comfortably between research and delivery — and between engineers and the business. That’s what I bring to a team.

  • Islamabad, Pakistan
  • PKT (UTC+5) · Remote-first
  • Lead · Manager · Senior IC
Portrait of Usman Malik

Usman Malik

AI Engineering Lead

Agentic AI LLM Cost −60% CV 4×
~8 yrs Applied AI
400+ Models
YC S25 Lead Eng

  • Find

    Spot the repetitive, high-volume work where AI clearly pays for itself — and the model that does it at the lowest cost.

  • Build

    Research, prototype and ship: agents, RAG pipelines, real-time vision — tested, monitored and production-grade.

  • Lead

    Build and grow the team around it — mentoring engineers, running research processes and AI workshops.

02Expertise

The full stack of applied AI.

From the data and the GPU to the agent and the business case — the toolkit I use to take AI from idea to production.

  • LLMs & Agents

    Agentic workflows, RAG and evaluation that hold up in production — not just in demos.

    • RAG
    • Agentic workflows
    • CrewAI
    • MCP
    • LLM evaluation
    • Prompt engineering
    • Fine-tuning
    • OpenAI Assistants API
    • GPT
    • Gemini
    • Llama
    • Mistral
    • DeepSeek

  • Computer Vision

    Real-time video analytics, from GPU servers to Jetson edge devices.

    • DeepStream
    • Triton
    • TensorRT
    • NVIDIA TAO
    • YOLO
    • SAM
    • VLMs
    • Pose estimation
    • OCR
    • OpenCV
    • Jetson

  • LLM Infra & Cloud

    Picking the best-value model and hosting for every workload.

    • Vertex AI
    • Azure
    • AWS Bedrock
    • S3
    • Ollama
    • vLLM
    • Docker
    • Jenkins CI/CD

  • Engineering

    The systems work around the models.

    • Python
    • C++
    • Django
    • Kafka
    • Redis
    • PyTorch
    • TensorFlow
    • Git/GitLab

  • MLOps & Data

    Reproducible training, tracking and data versioning.

    • DVC
    • Weights & Biases
    • ClearML
    • Unit & E2E tests for AI
    • Synthetic data
    • Annotation pipelines

  • Leadership

    Building teams, research processes and AI literacy.

    • Team leadership
    • Mentoring
    • AI research process
    • AI workshops
    • Responsible AI
    • Client reporting

03Experience

Where I’ve shipped.

Eight years across product companies, enterprise and a YC startup — always between research and delivery.

Enghouse Interactive

May 2024 — Present · Remote

  1. Senior Software Engineer · AI/ML R&D

    • Lead LLM R&D — benchmark closed and open-source models (Gemini, GPT, Llama, Mistral, DeepSeek) on Vertex AI, Azure, AWS and locally via Ollama/vLLM, cutting model costs 50–60% across several use cases.
    • Build agentic assistants and RAG pipelines that give customer service agents accurate, context-aware answers from knowledge bases.
    • Designed unit and end-to-end tests plus an automated evaluation framework for LLM services; set up Jenkins CI/CD across GCP, Azure and AWS.
    • Built pose-estimation patient fall detection deployed on Jetson edge devices in hospital rooms.
    • Lead a two-person ML team, build internal AI tools for the engineering org and run workshops on responsible AI usage.
  • RAG
  • Agents
  • Vertex AI
  • vLLM
  • Ollama
  • Jenkins
  • Jetson
View full résumé

04Selected work

Things I’ve built.

Case studies from production systems. Client work is described at a generalised level — happy to go deeper in conversation.

  1. 01 LLM & Agents · Enghouse Interactive

    Best-Value LLM Benchmarking

    A benchmarking and evaluation programme that matches every LLM use case with the most cost-effective model — closed or open source, cloud or local.

    • 50–60% reduction in model costs across several use cases
    • Objective, repeatable model selection instead of guesswork
    • Vertex AI
    • Gemini
    • GPT
    • Llama
    • Mistral
    • DeepSeek
    Read case study

  2. 02 Computer Vision · DP World

    Video Analytics at 4× Throughput

    Re-architected real-time camera pipelines from Triton to DeepStream, quadrupling inference throughput per camera on the same hardware.

    • 4× inference throughput per camera
    • More streams per GPU — lower hardware cost per camera
    • DeepStream
    • TensorRT
    • Triton
    • C++
    • Kafka
    • Redis
    Read case study

  3. 03 LLM & Agents · Enghouse Interactive

    Agentic Knowledge Assistant

    Agentic assistants and RAG pipelines that give customer service agents accurate, context-aware answers from large knowledge bases.

    • Faster, more accurate responses for customer service teams
    • Answer quality measured automatically — not by spot checks
    • RAG
    • LLM agents
    • OpenAI Assistants API
    • Code interpreter
    • Jenkins
    • GCP
    Read case study

  4. 04 LLM & Agents · Risely (YC S25)

    Agentic University Operations

    Led backend engineering for a YC-backed agentic AI platform that automates university operations — with staff approving key actions.

    • Agent workflows grounded in live institutional data
    • Safe automation with staff approval gates
    • Django
    • CrewAI
    • MCP
    • Python
    • Prompt optimisation
    Read case study

  5. 05 Computer Vision · Healthcare deployment

    Patient Fall Detection on the Edge

    Pose-estimation fall detection running in real time on NVIDIA Jetson devices inside hospital rooms.

    • Real-time alerts at the bedside
    • Privacy-preserving — video never leaves the room
    • Pose estimation
    • NVIDIA Jetson
    • TensorRT
    • DeepStream
    • PyTorch
    Read case study

  6. 06 Computer Vision · DP World

    On-Prem VLMs for Fire & Smoke

    Deployed and evaluated Llama 90B vision-language models on-premise to detect fire and smoke in scenes where classic detectors struggle.

    • Better coverage of hard edge cases
    • Fully on-prem — no footage leaves the site
    • Llama 90B Vision
    • VLMs
    • On-prem GPUs
    • Python
    • Prompt engineering
    Read case study

More work

View full project & open-source archive

Interactive Estimator

See what engineering rigor saves.

Modelled on production outcomes: 50–60% LLM cost reduction via benchmark-driven model routing (Enghouse Interactive) and 4× camera throughput via Triton → DeepStream migration (DP World).

01 · LLM & Agentic Workloads

Best-Value Model Routing

Optimised monthly spend \$5,400
Estimated annual savings \$79,200 / yr

Strategy: automated evals + routing across Gemini, GPT, Llama, Mistral, DeepSeek & local vLLM/Ollama.

02 · Computer Vision Infrastructure

DeepStream + TensorRT 4× Density

Streams per GPU 4× higher
Hardware footprint saved 12 fewer GPUs (−75%)

Strategy: model retraining + TensorRT conversion + custom C++ DeepStream probes streaming to Kafka/Redis.

Open source

On GitHub

@uma291

Research I study & build on

05Leadership & Hiring

I lead teams that ship.

Whether you’re hiring an engineering leader or looking to join a high-output AI team — here is how I operate, mentor and build.

  • Start from the business case

    Before writing code I find the repetitive, high-volume work where AI clearly pays for itself — and say no to the rest.

  • Measure, don’t guess

    Benchmarks and automated evals choose the model, not hype. That discipline is how we cut LLM costs 50–60%.

  • Grow the people

    I mentor juniors into owners, run AI workshops and build research processes others can follow without me.

  • Ship safely

    Tests for LLM services, CI/CD, data masking and human-in-the-loop approvals — AI the business can trust.

Leadership track record

  1. 2018 — 2020

    Built & led the annotation team

    DARVIS

    Designed the data pipeline end to end — acquisition to delivery — with remote tooling on AWS and dataset version control.

  2. 2021 — 2023

    Led computer vision engineers

    DARVIS

    Guided junior and mid-level engineers through real-time detection, tracking and pose-estimation client deployments.

  3. 2024 — Now

    Lead a two-person ML team

    Enghouse Interactive

    Own LLM R&D direction, internal AI tools and responsible-AI workshops for the wider engineering team.

  4. 2025

    Led backend engineering

    Risely (YC S25)

    Owned backend architecture and database design for an agentic platform at a YC-backed startup.

06 — What’s next?

Let’s build AI that pays for itself.

Hiring for an AI lead, engineering manager or senior ML engineer? Looking to cut LLM/CV infrastructure costs or join a high-output AI team? My inbox is open — I usually reply within a day.

Say hello
Quick inquiry builder Opens pre-filled in your mail client or copies to clipboard