whoami

Hamza Shergill

Hamza Shergill

Stop losing time to manual processes. I build AI systems that solve operational problems. Not just moving data, but thinking, deciding, and acting on your behalf.

Available for fractional engagements · Birmingham, UK · Remote

03Production agent systems shippedMateOS · Workmancer · Kyrios
07Engagements completedupwork.com/freelancers
$3K+Delivered via Upworkindependent of direct clients
runtime-log

What these systems log in production

tail -f agent-events.log streaming

Simulated: a representative sample of the events MateOS, Kyrios, and Workmancer emit while running, replayed here on a loop.

featured-systems

Agent systems I've shipped

Production agentic systems built for operators who needed a workflow to stop depending on a human being available. Each one designed, built, and handed over to run without babysitting.

01

MateOS

flagship open-source ↗ shipped

full build · independent CEO client · autonomous scheduling agent

A CEO was losing hours a week to appointment setting and call screening. MateOS takes inbound contact by call, email, or chat, reads intent, checks live calendar and availability rules, and books, reschedules, or follows up. No human in the loop. Every booking syncs to CRM and calendar, with fallback logic for no-shows and unqualified inquiries.

Result: the client no longer touches scheduling or call screening at all; every meeting on the calendar arrives pre-qualified and confirmed.

AI Agent Development CRM Automation API Integration Prompt Engineering
02

Workmancer

flagship live ↗ shipped

orchestration layer · Workmancer · multi-agent “AI employee” platform

Workmancer deploys “AI Employees” that get assigned real work. The hard problem was making them collaborate across departments the way human teams do, instead of working in isolated silos. I built the task orchestration and cross-department communication layer: each agent runs its own role, context, and task queue, but shares a common protocol to request help, pass context, and report completion.

Result: clients deploy a full “workforce” of AI Employees that completes multi-department tasks end to end, with no human routing work between them.

Multi-Agent Systems Business Process Automation API Integration Prompt Engineering
03

Kyrios

flagship live ↗ shipped

end-to-end build · LeadHunt INC · autonomous outreach system

LeadHunt INC needed to offer clients a fully automated outbound system with no manual SDR team running campaigns. Kyrios sources leads against a client's ICP, validates and scores them for fit and data accuracy, then runs outreach autonomously across Email, LinkedIn, and cold voice, each channel with its own messaging, follow-up, and objection-handling logic feeding back into a shared qualification loop.

Result: clients run full outbound campaigns, sourcing through booked call, on autopilot across three channels at once.

AI Agent Development Lead Generation CRM Automation Prompt Engineering
side-builds

More things I've shipped

Independent products built outside client work: the same instinct for turning a manual process into a system that runs itself, applied to my own ideas.

capabilities

What I actually build

Not prompts in a chat window, but systems with state, memory of what happened, and a clear line between what the agent decides and what it hands to a human.

01

Agent Architecture & Orchestration

Multi-agent systems where each agent owns a role, a context window, and a task queue, with a shared protocol for handing work to each other, the way Workmancer's AI Employees collaborate across departments.

02

Tool Use & System Integration

Agents that act, not just answer: wired into calendars, CRMs, and messaging channels through direct API integration so a decision actually executes. MateOS books the meeting; it doesn't just suggest a time.

03

Reliability & Handover

Every system ships with fallback logic and explicit handover conditions (for no-shows, unqualified leads, anything the agent shouldn't decide alone) so it survives contact with the real world, not just the demo.

ai-engineer-skillset

My Core Expertise as AI Engineer

The full stack behind every system on this page, from how a model is prompted up to the plumbing that keeps an agent reliable once it's live.

01

LLM Foundations & Prompt Engineering

Structuring prompts, system messages, and context windows so a model behaves reliably, not just once. Few-shot examples, reasoning scaffolds, and output formatting that downstream code can actually parse.

02

Agentic Systems & Tool Use

Giving a model tools it can call (calendars, CRMs, search, code execution) and a loop that lets it plan, act, observe, and decide what to do next: the core pattern behind MateOS and Kyrios.

03

Multi-Agent Orchestration

Coordinating several specialized agents that hand off work, share context, and report completion back to a shared protocol, the pattern Workmancer's AI Employees run on.

04

RAG & Knowledge Retrieval

Chunking, embedding, and retrieving the right context from a knowledge base so an agent answers from real data instead of guessing, with hybrid search when keyword matching still matters.

05

Model & API Integration

Working directly with the Anthropic Claude API and OpenAI API: streaming responses, function calling, rate limits, and keeping token costs predictable once a system is in production.

06

Workflow Automation

n8n, Make, and Zapier for the plumbing every agent needs around it: triggers, webhooks, scheduled jobs, and connecting a dozen SaaS tools without writing a backend from scratch.

07

Evaluation, Guardrails & Fallbacks

Testing what an agent actually does against what it's supposed to do, then building the explicit fallback and handover logic for the cases it shouldn't be trusted to decide alone.

08

Data Pipelines & Python

Python for the glue code: web scraping, REST API integration, data cleaning, and the scripts that keep an agent's inputs accurate and its outputs structured.

build-pipeline

How an autonomous system gets built

The full lifecycle a system goes through before it's trusted to run on auto, security and evaluation included, not bolted on after launch.

01Discover

scope

Audit the manual process, find exactly where a human is the bottleneck, and define what “running on auto” should actually mean for this workflow.

02Design

guardrails

Map the decision points: what an agent can safely decide on its own vs. what always routes to a human, and design the guardrails around each one.

03Build

integration

Wire the agent logic into the real tools it needs: APIs, CRM, calendar, comms channels, and the internal data it has to reason over.

04Secure

security

Scope every credential to least privilege, sandbox tool calls, add prompt-injection and data-leak defenses, and log every action the agent takes for audit.

05Evaluate

eval suite

Build a test set of real and adversarial cases before launch, then score accuracy, latency, and cost against it, not just “it worked when I tried it.”

06Deploy

staged rollout

Ship behind a staged rollout with a fallback path for every failure mode and a human-approval gate on anything high-stakes or irreversible.

07Monitor

observability

Watch the first real runs closely: logs, alerts, and drift checks on both the model's behavior and the systems it touches.

08Operate on Auto

autonomous

Hand over full autonomy once trust is earned, with a kill-switch, scheduled re-evaluation, and clear ownership for when it needs a human again.

automation-and-growth-systems

Where the agent instincts came from

Before agent orchestration, I was already building the conditional logic and sequencing that make a funnel run itself: tagging, triggers, branching sequences. That's the same muscle multi-agent systems use; the funnels just taught it to me first.

Auto Sales Funnel: Low-Ticket Offer

Built the trigger and tagging logic behind a self-running low-ticket funnel: cold traffic converts without a human touching an individual lead.

Master Funnel: Micro Offer → High-Ticket

Designed a multi-stage sequence that automatically escalates micro-offer buyers into high-ticket conversations based on engagement signals, not guesswork.

Lead Magnet Automation: My Mummy Australia

Built the opt-in, tagging, and nurture-sequence automation behind a lead-magnet funnel for a targeted consumer audience.

system-status

Stack

AI & Agents

LLM Application Development, AI Agent Development, Multi-Agent Systems, Prompt Engineering, OpenAI API, Anthropic Claude API, Agent Orchestration

Automation

n8n, Make (Integromat), Zapier, GoHighLevel, Workflow & Business Process Automation, CRM Automation, API Integration

Development

Python, JavaScript, REST API Integration, Web Scraping & Data Pipelines, Chatbot Development

Go-To-Market

Lead Generation & Scoring, Cold Outreach Systems, Marketing Automation, Growth & Funnel Strategy

background

About

I spent years running marketing and content functions before moving into engineering. That means I don't just build systems that technically work; I build them to hit a business number someone is actually accountable for.

These days that means agentic AI systems: agents that qualify leads, run outreach, manage scheduling, and coordinate with each other the way a real team would, built on n8n, Make, Python, and direct LLM APIs. The marketing background stays useful: it's the difference between shipping a working pipeline and shipping one that moves the number the client is being measured on.

  • Chief Marketing OfficerAetmaad.com
  • Creative Marketing DirectorUber
  • Digital Marketing ExecutiveDigital Ink Marketing
  • Content Leade-Baldi
Education
BEng, Industrial Engineering
University of Engineering & Technology, Lahore · 2014-2018
get-in-touch

Contact

Available for fractional AI engineering engagements, usually starting with a short scoping call.