I build systems that read thousands of documents so people don’t have to.
A year shipping production AI at AIVSTUDIOS: pipelines for legal record building, clinical summarization, and platforms that run themselves.
01Selected projects
Every project below runs against real data for real users, across legal, healthcare, sports and commerce.
Flagship project
The problemLegal teams were manually collecting, reading and reformatting thousands of court PDFs into standardized records. Slow, repetitive, and easy to get wrong.
What I builtAn automation bot pulls the documents in. One agent reads each court decision, a second classifies it, then an automation layer handles the shrinking, standardizing, numbering, titling, splitting, duplicate checks, record building and bookmark generation.
Then the humans step inA custom PDF editor lets users refine any output, with agents behind it handling edits and rendering, plus automation for volume assembly, OCR and final flattening.
Lead developer · Python, Multi-agent LLM, OCR, PDF engine, Workflow automation
Production systemsShipped for clients and running live8
Legal
Runs every week once new court decisions land. Pulls case data across four departments, then agents build arguments, consolidate them and assemble a finished newsletter served over an API. Built end to end by me.
PythonFastAPIAgentsScheduling
Legal
Reads the record on appeal alongside the appellant and respondent briefs and the reply brief. One agent builds the supporting and counter arguments, a second consolidates them into a single line of reasoning, and a third drafts the questions a judge would put to each side.
PythonMulti-agent LLMLong-context RAGPDF
Health
Users upload very large patient record PDFs; multiple agents condense them into short, organized summaries doctors and patients can actually read. I built a backend module of the pipeline.
OpenAIGeminiPythonPDF
Sports
Court assignment, live rankings and player statistics for clubs in Canada, with full authentication, AI-based player evaluation and an AI coach that guides players on what to work on next.
PythonAuthAI evaluationCoaching agent
Commerce
Shopping store backend and companion app showing real-time user activity and live ordering statistics as they happen.
BackendMobile appRealtime
Ops
An end-to-end invoice build-and-send flow as a multi-step, multi-zap workflow wired into ActiveCampaign and QuickBooks.
ZapierActiveCampaignQuickBooks
Legal
Automated record assembly for a US legal use case, sourcing case data through the CourtListener API.
PythonCourtListener API
Vision
Attendance system with live tracking built on AdaFace. It recognizes people as they arrive and logs it without anyone touching a device.
AdaFaceComputer visionPython
Independent buildsProducts started and finished on my own5
Independent
An internship platform where the testing, goal assignment and evaluation are all handled by AI, so companies can run a structured program without running it manually.
AI evaluationPythonWeb
Independent
AI-driven parking management: allocation and day-to-day operations for parking spaces.
AIPython
Independent
Local wholesalers were losing track of inventory and order bookers. I built the full management system plus a Flutter app so order bookers can work from the field.
FlutterPythonPostgreSQL
Independent
Scrapes and organizes leads, checks whether each business is real and active, then feeds automated outreach.
ScrapingValidationEmail automation
Independent
A storefront built for search visibility from the ground up rather than patched on afterwards.
Next.jsPythonPostgreSQL
Research and courseworkWhere the fundamentals were built2
Final-year project
A modular, privacy-first AI assistant that works online and offline. Understands voice and text commands for productivity, entertainment and system control, with local speech recognition and a local knowledge base so it keeps working without a connection.
FastAPIRasaWhisperVoskVector DB
Academic
Bike-sharing demand prediction, spam email detection, house price estimation, an e-commerce recommender and customer churn prediction.
scikit-learnNumPyNLP
02ExpertiseDaily drivers
Chosen for what ships and stays running, not for what’s new this month.
03ExperienceIslamabad, Pakistan
A year of production AI work on top of a computer science degree spent on machine learning, vision and automation.
Jul 2025 to Present
Building multi-agent pipelines and AI platforms for clients across legal, healthcare, sports and commerce. Lead developer on the document customization pipeline.
2025
26th globally in the opening stage at over 80% accuracy; 57th globally in the final stage.
2021 to 2025
Machine learning, computer vision and automation throughout. Final-year project: RJ Assistant, an offline-capable AI assistant.
04ContactOpen to work
Tell me what your team is doing by hand that a pipeline should be doing instead. I’ll tell you honestly whether it’s worth building.