Claude Certified Architect · Lahore, Pakistan · UTC+5
Muhammad Hamza Khalid
IhelpSMBsautomatetheiroperationswithClaudeandMCP.
Custom AI agents, MCP server integrations, and Claude-powered automation that removes manual, repetitive work — so your team stops doing data entry and starts doing their actual jobs.
Open to remote · SMB-focused · No research, just automation that ships
- Users served
- 7,000+
- Bookings / day
- 20–30
- Spend reduction
- ~80%
- Note agreement
- 88%
Talk to my digital twin01 / About
How I work
Hamza Khalid is a Claude Certified Architect who helps small and mid-size businesses replace manual, repetitive work with reliable AI automation. The work that eats hours every week — copying data between tools, drafting the same reports — gets done without anyone touching it.
Your team stops pasting numbers into spreadsheets and retyping customer details. What used to take three hours takes five minutes, copy-paste errors disappear, and people get back to the parts of the job that actually need them.
- Based
- Lahore, Pakistan ·
- Availability
- Open to remote
- Experience
- 3+ years engineering
- Certified
- Claude Certified Architect
02 / Process
How it works
Step 01
Find the expensive work
I time the jobs that eat your week, then we pick the one with a payback you can actually calculate.
One client was spending 35 minutes typing in every order. That is the kind of number worth automating.
Step 02
Connect it to your tools
It works inside the tools you already pay for, each opened only as far as it needs. Nothing is copied out to another company's system.
It can read the thread and draft the reply, but it cannot send on your behalf unless you decide it should.
Step 03
Let it run without you
Work starts itself — on a schedule, or the moment an order arrives. Nobody has to remember to run it, and nobody has to learn a new tool.
Rebuilding how one client's jobs were run cut the cost of running them by about 80%.
Step 04
Prove it keeps working
Every change is checked against real cases before it goes near your customers, and every run is recorded — so drift shows up early.
This is the part most people skip. It is the reason these keep working months later.
03 / Systems
SystemsI'vebuilt
A few highlights — not the full list.
Multi-clinic veterinary group
7,000+ people served, 20–30 bookings a day
WhatsApp booking assistant
Every appointment had to go through the front desk, so bookings stopped whenever the phone did. Patients now book themselves over WhatsApp, and it hands over to a person when it should.
Enterprise ERP
Multi-day processes down to minutes
One shared way in to the ERP
Every new automation meant rebuilding the same connection to their ERP from scratch. One shared layer replaced that, with each automation given access only to the parts it needs.
Recurring work
~80% lower running cost
Jobs that run themselves, cheaply
Routine jobs paid full model cost on every run and still needed someone to start them. Rebuilt to fire on a schedule inside a locked-down sandbox, so they run unattended for a fraction of the cost.
50+ orders a day
35–40 minutes down to under 2, per order
Reading orders automatically
Orders arrived as free text and were retyped by hand. A model tuned on their own orders now pulls the details out, and anything it is unsure about goes to a person. About 99% need no correction.
Knowledge and memory
Answers from the company's own handbook
Looking things up on demand
The assistant needed the company's own documentation, and long conversations eventually stopped fitting. It now looks things up when needed instead of carrying everything along.
Testing and monitoring
Faults caught before release, not by customers
Proving it still works
A reworded prompt could quietly break how the assistant picked its tools, with no sign until someone hit it in production. Every change now runs against real cases first, and every run is recorded.
Earlier backend systems — real-time clinical transcription and summarisation, and an event-driven trading integration at 10,000+ concurrent users — are detailed under Experience.
04 / Experience
Experience
Feb 2025 — Present
AI Engineer
Cogent LabsCurrent- Leads a team of 4 engineers on client AI engagements end-to-end — discovery, architecture, and delivery.
- Builds production agent systems: multi-agent orchestration, MCP tool architecture, retrieval pipelines, and evaluation harnesses.
- Runs company-wide workshops on MCP and Agent Skills.
LangGraphpydantic-aiMCPFastMCPpgvectorDeepEvalLangfusePythonMay 2024 — Feb 2025
Software Engineer
Cogent Labs- Built a real-time transcription and summarisation pipeline reaching 88% agreement with clinician notes under LLM-judged evaluation.
- Built an event-driven integration layer on RabbitMQ and Polygon webhooks for a trading platform serving 10,000+ concurrent users.
- Cut database load 65% and query response time 80% through Redis caching, PostgreSQL read replicas, and hand-tuned SQL on TimescaleDB.
PythonRabbitMQRedisPostgreSQLTimescaleDBPolygonJun 2023 — Feb 2024
Junior Software Engineer
Apex Space- Built Django and PostgreSQL backend services for catalog, checkout, and order processing across e-commerce clients.
- Built React front-end features against the same REST APIs.
DjangoPostgreSQLReactREST
B.Sc. Computer Science — University of Engineering and Technology, Lahore
Graduated May 2024
05 / Contact
Getintouch
Open to remote roles and contract work on production agent systems. Email is the fastest route.
© 2026 Muhammad Hamza Khalid
Lahore, Pakistan · UTC+5