AI and Full Stack Engineer · Lahore, Pakistan
Django and Go backends at 10K requests per second. React and Next.js frontends, agentic LLM systems in production.
Full time roles · Contract builds · White label delivery for agencies
Selected work
Task queues, regulated SaaS, live markets, device fleets, AI agents. Six in full below, six more after them. Every card opens a case study, most open the live product too.
01 / 12 · Open source
32.8 times Celery throughput.
Distributed task queue with a Rust core exposed to Python through PyO3. A drop in alternative to Celery that runs async, blocking and CPU bound tasks in one process.
Rust · PyO3 · Redis 7
02 / 12 · AI agents in production
10K requests per second. 50K live sockets.
Unified Facebook, Instagram and WhatsApp inbox on the Meta Graph API, with LLM agents doing auto response, summarisation and intent routing.
Django Channels · Go · LangGraph
03 / 12 · Regulated SaaS
500K lines of regulated Django.
Compliance critical drug and alcohol testing platform for Canadian administrators, labs and medical review officers.
Django · Vue · Celery · AWS
04 / 12 · Live markets
Atomic settlement on a double entry ledger.
Prediction market platform. Order book, double entry ledger, wallets and settlement, with live market data pushed over WebSockets.
Django · PostgreSQL · Next.js · WebSockets
05 / 12 · Device automation
A fleet of Android devices, driven from one screen.
Desktop and web platform for orchestrating fleets of connected Android devices, with a visual graph builder for automation workflows and live mirroring.
Django · Node.js · Appium · scrcpy
06 / 12 · On PyPI
Pixel faithful DOCX to HTML. Zero dependencies.
Pure Python OOXML rendering engine. No LibreOffice, no Word. Tables, shapes, floating images and fonts inlined into one self contained HTML file. Replaced a paid .NET dependency in production.
Python · standard library only
Also shipped
SEO platform
Trusted by 6,000+ SEOs.
Case study
Marketplace
A two sided marketplace, both flows shipped.
Case study
Staffing ops
Bulk comms to hundreds at once.
Case study
Trading platform
Multitenant trading core, wallets to bots.
Case study
WhatsApp commerce
Receipt forensics before fulfilment.
Case study
AI agent
Decisions pulled out of Slack noise.
Case study
The positioning
"I own the whole path. Schema to socket to screen, then the box it runs on. I take the on call, I make the architecture call, and I do not need hand holding to do either."
Five years of production systems where the failure mode is part of the design, not an afterthought. Django, DRF and Go backends, React and Next.js frontends, agentic LLM systems on LangGraph and CrewAI, and self hosted image and video generation on GPU infrastructure.
The work style is direct: understand the product, find the bottleneck, build the system, measure the result, and leave the codebase easier to operate than I found it.
10K
requests per second at peak on the Django and Go core behind ChatsX
50K
concurrent WebSocket connections on a rebuilt Django Channels layer
300K
background jobs through fault tolerant pipelines with dead letter queues
32.8×
Celery throughput from Cauli, the open source task queue I author
Three ways to work with me
Hiring managers, founders and agencies all land here. Each of you needs something different from me, so here is the honest version of each.
Full time or long term embedded, remote from UTC+5. Backend ownership at real traffic, the architecture calls, the on call rotation, and mentoring the juniors through review. Five to six hours of overlap with the UK and EU, four with US East.
Senior full stack · AI systems · Remote
Recruiter snapshotContract work with a scope and a finish line. Frontend, backend, QA, deploy and infra, plus the scaling work after it ships. You get one accountable engineer, not a handoff chain, and a codebase your next hire can read.
Fixed milestones · Weekly demos · Your repo
Send a briefWhite label delivery. I build under your brand, in your repo, to your standards. You own the client relationship and the credit. I never appear in front of your client unless you want me to, and nothing here ever names them.
Your brand · Your process · Quiet
How white label worksNumbers from production
API latency
900ms 220ms
p95 on the Django and Go core at 10K requests per second. Query rewrites, multi layer caching, PgBouncer pooling.
Realtime layer
5,000ms 100ms
Django Channels rebuilt for 50K concurrent sockets. A 96 percent cut in response time.
Throughput
32.8× Celery
Cauli saturates the host instead of the framework. 27,060 tasks per second on 4.71 worker cores.
Under 3 days
Reverse engineered a dead vendor WebSocket protocol and shipped a replacement client, restoring a revenue critical integration.
300K jobs
Through fault tolerant pipelines with dead letter queues, idempotent retries and priority routing.
25 to 7 min
Deploy time, cut with Azure DevOps pipelines and containerised builds.
RCE closed
Found and fixed a Django remote code execution hole before it ever reached production.
Technical depth
I have reverse engineered a deprecated vendor WebSocket protocol into a working Python client in under three days, closed a Django remote code execution hole before release, written an Office Open XML renderer in pure Python, replaced managed cloud load balancing with a self built Caddy stack, and put FLUX and LTX video generation on RunPod GPUs to kill a recurring inference bill.
What makes the work different
Measure before changing
Latency, query time, job failures and delivery speed get tracked against real outcomes, and the before number is kept.
Full product ownership
No ticket honeymoon. I take the problem, not the ticket: frontend, backend, QA, deploy and infra, then the scaling after it ships.
AI assisted delivery
Claude Code in the loop. Subagents, git worktrees, custom workflows and QA driven through MCP. Leverage, not a replacement for judgment.
Security as product survival
RCE risk, tenant isolation, permissions and integration boundaries get handled before they become incidents.
Writing
Long form pieces on the parts that fought back. Parsers, task queues, and benchmarks that argue with their own conclusions.
5 July 2026 · 6 min read
How a simple .docx to HTML task turned into writing an Office Open XML rendering engine from scratch, and what I learned about fonts, painting order, and measuring perfect.
Python · OOXML · Open source
1 July 2026 · 9 min read
FastAPI, Litestar, Django Ninja, DRF and Django Bolt. Same query, same database, same load. The framework was rarely the bottleneck.
Benchmarks · Python · PostgreSQL
Send the product, stack, deadline, current bottleneck, and what success should look like. I will reply with a concrete technical next step, whether you are hiring, contracting, or white labelling.
