● Pre-med → AI · Lahore, Pakistan
I was supposed to be a doctor.
Instead I build AI.
I build AI systems that run businesses on autopilot, then show exactly how I did it. I'm Shahzaib, an AI automation engineer at Automaxion shipping production agents, voice bots, and pipelines for clients across three continents. Building in public, one system at a time.
Building in public
I build it, then I show you how.
Live products I ship on the side, and the breakdowns I post as I go.
Hisaab
liveA personal finance tracker that answers the only question that matters: where does the money actually go?
VisitToolbelt
liveFree, privacy-first browser utilities for PDF, image, and dev work that run entirely on your machine. No uploads, no sign-ups.
VisitPromptly
buildingA faster way to run, organize, and reuse your best AI prompts. In progress.
Latest, in public
I post every build on LinkedIn. Follow along
The Path Not Taken
Two careers.
One person.
A five-month decision in between.
Move to Bahawalpur
FSc pre-medical · biology, chemistry, physics
182018—
FSc complete
Pre-med track. Frogs dissected, organic reactions memorized.
2020205–6 months of research
Reading everything I could find about what to do next.
The fork
Pre-med future vs a degree no one in my town had heard of.
212021BS Artificial Intelligence, IUB
Started from zero. Classmates had been coding since high school.
—
222022Ubuntu, daily, for 2 years
A senior recommended it. It broke things; I fixed them. Stopped fearing the terminal.
—
232023Neurafinity Club, co-founded
Plus the first of four Python bootcamps. Teaching forced understanding.
—
242024TA for the AI certificate course
Taught the second batch of the same course I’d just finished: ML, deep learning, computer vision.
—
252025Hired by Automaxion · 7th semester
Before I’d graduated. Production AI for real clients.
—
262026Graduated · CGPA 3.65
Outperformed peers who’d been coding since school. Still rebuilding LinkedIn from zero.
I didn't inherit this path. I researched it for five months and chose it.
Projects
Built in the open.
The code for these is public. Read it, run it, fork it.
- 01MIT
Telegram Order Agent
AI order-taking for restaurants. The AI never touches money.
Customers order through Telegram in English, Roman Urdu or a mix. An AI agent matches what they say to the real menu, collects delivery details and shows a cart they confirm with a button. The interesting part is where the rules live: prices, minimums, delivery fees and order states are enforced in Postgres, not in the prompt, so swapping the model changes nothing about correctness.
- Orders are placed only by a button press, never by typing "ok"
- Live kitchen board that messages the customer on every status change
- The agent has exactly seven database tools and cannot write its own SQL
n8nPostgresNext.jsTelegram Bot API - 02MIT
YTDM
A fast, modern download manager for YouTube.
A desktop app in the style of IDM, built for YouTube. Paste a video or a playlist, pick a quality, and it downloads in a managed queue that survives crashes and restarts. Its download engine updates itself independently, so YouTube changes do not break it.
- Queue with pause, resume, retry and crash recovery
- Playlists, subtitles, quality presets and a bandwidth cap
- Self-updating app and engine, builds produced by GitHub Actions
ElectronTypeScriptyt-dlpFFmpeg - 03MIT
Memory-Mate
An AI chat assistant that remembers you between sessions.
A chat assistant with persistent, per-user memory. It keeps the important facts from each conversation and a running summary, stored in SQLite, so the next session starts with context instead of from zero.
- User-specific long-term memory and conversation summaries
- Simple username sign-in, responsive chat interface
PythonAgnoStreamlitSQLite - 04FYP, public repo
AI Surrogate
Final year project: an assistant that acts, with you approving.
An agentic mobile assistant built for the BS Artificial Intelligence final year project at IUB. It chats with memory, drafts Gmail replies that you edit and approve before they send, and schedules meetings through conversation, in English, Urdu and Punjabi.
- Human approval step before anything leaves the app
- Voice input and spoken replies
FastAPIReact NativeMistralOpenAIPostgres
Production systems at Automaxion
Built for clients, so named by category only.
- Autonomous social media systemMake.com, Airtable, LLM APIs
- Cold outreach pipelineAirtable, Make.com, Instantly
- Data enrichment dashboardNext.js, custom APIs
- Psychological assessment automationCustom scoring, PDF reports, ActiveCampaign
- AI teaching assistant for long documentsElevenLabs narration, quizzes
- Self-hosted infrastructure for all of the aboveLinux, Docker, Nginx
What I work in
Six clusters.
Each one rhymes with something I almost learned.
Automation
Like building a treatment protocol. Same input, predictable outcome.
- ·n8n (self-hosted)
- ·Make.com
- ·Zapier
- ·Airtable
- ·Instantly
- ·ActiveCampaign
AI agents & LLMs
Differential diagnosis at scale. Ask the right questions, narrow the answers.
- ·OpenAI
- ·Claude
- ·Mistral
- ·CrewAI
- ·Agno
- ·Function calling
Voice AI
The bedside manner. Tone and timing matter more than the words.
- ·VAPI
- ·Retell
- ·Bland AI
- ·ElevenLabs
Engineering
Anatomy of a system: every part connects, nothing is decoration.
- ·Python
- ·TypeScript
- ·Next.js
- ·React
- ·Node
- ·Postgres
Infrastructure
Like running your own pharmacy. You stock it, you maintain it, you own it.
- ·Linux / Ubuntu
- ·Docker
- ·Nginx
- ·Self-hosted VPS
- ·Hetzner
- ·DigitalOcean
Teaching
Residency rounds: you learn it by teaching it to someone newer.
- ·4 Python bootcamps
- ·AI certificate TA
- ·Neurafinity workshops
- ·YouTube lectures
What I taught while I learned it
Teaching forced me to actually understand things.
I co-founded the Neurafinity Club at IUB, ran four online Python bootcamps for juniors and students, and served as teaching assistant for the second batch of a 10-month AI certificate course, the same course I had just completed in the first batch.
You can't explain what you don't know. So I went deep on the parts I needed to teach: classical ML, deep learning, computer vision, then later automation and agentic AI. The bootcamps had no marketing. Students kept signing up because the previous cohorts told them to.
I'm still teaching. Most of the writing on /blog is the stuff I wish someone had explained to me when I was figuring it out.
4
Python bootcamps taught
10mo
AI certificate course TA
3.65
CGPA, BS AI · IUB
PM
Laptop Scheme awardee
What I build
The systems I ship, in public.
Three things I do well. I build them for clients, and I break them down so you can see how. If your problem rhymes with one of them, email me.
- 01
Workflow Automation
n8n, Make.com, Zapier. End-to-end pipelines that actually run.
- 02
AI Voice Agents
VAPI, Retell, ElevenLabs. Inbound, outbound, in your stack.
- 03
Custom AI Applications
Next.js + Python + LLM APIs. Built from scratch when no platform fits.
What I'm open to
Scholarships.
Research collaboration.
Freelance.
A good email.
I read every message. Email is best. If you're a scholarship reviewer, recruiter, or want to teach together, say hi.