Sheraz Labs

From chatbots to pipelines - AI that actually does the work.

I'm Sheraz, an AI engineer in London. I design, build and deploy agentic systems, from first sketch to production on AWS.

See my work Let's talk

The foundations behind the builds

  • AWS Solutions Architect – Associate SAA-C03 · 2026
  • Google Project Management Professional Certificate · Coursera
  • Google Data Analytics Professional Certificate · Coursera
  • MSc Data Science Kingston University London
  • BS Computer Science University of Sialkot

The stack I ship with

  • LangGraph
  • LangChain
  • Gemini
  • Groq
  • DeepSeek
  • Whisper
  • scikit-learn
  • TensorFlow
  • OpenCV
  • pandas
  • NumPy
  • Python
  • Django
  • PostgreSQL
  • AWS
  • Docker
  • Nginx
  • Linux
  • Git

Case studies

Experiments and research: what AI can do when it is built properly.

Video Editing Agent

Research prototype

The problem

Captioning and overlaying a video means learning a timeline editor. The Video Editing Agent replaces the timeline with a chat: upload footage, describe the change, and a LangGraph agent transcribes, captions, places overlays and renders the file.

Hard problems solved

  • What's drawn matches what's reported. One layout pass feeds both rendering and the agent's view of the frame, so it can't describe a caption somewhere it isn't.
  • Concurrent agent writes. Moved off SQLite to Postgres so per-call token accounting stops queueing behind a whole-file write lock.
  • No-signup access, safely. Guest identities own jobs like accounts do, with timed retention that clears video first and chat later.
Read case study Try the research demo

C4 · Container view

C4 container diagram: nginx in front of a Django and LangGraph web app, with Postgres, media storage, Certbot, and external LLM and transcription APIs

Click to open full size

Blog

Build logs and bug war stories.

First posts coming soon

Write-ups of real problems from building these systems are on the way.

About

Sheraz Ali

Sheraz Ali

Agentic AI engineer, London

I'm a systems thinker based in London. I started out running a small electronics store while finishing a CS degree, built a loan platform in Django, and now design agentic AI systems end to end - from the graph the agent runs on to the servers it's deployed to.

I care most about the parts that decide whether an AI system survives real users: clear tools, costs that can't run away, and data that's deleted when it should be.

More about me

Working on an AI system? Let's talk.

Whether it's an agent that needs to reach production or a manual process that should be automated, tell me about it.

Email me