LLM Fine-Tuning
Fine-tuning & optimizing Llama, Mistral, DeepSeek and BERT for real accuracy — measurable loss and perplexity gains on your domain data.
Available for new projects
AI / ML & Data Engineer —
I'm an AI/ML & Data Engineer. I fine-tune LLMs, ship computer-vision systems, and build the data pipelines that feed them — plus RAG chatbots and workflow automation. I've also delivered production voice agents for clients. 100% Job Success on Upwork, every contract rated 5.0.
01 — About
I'm Muhammad Ali Mahmood — an AI/ML & Data Engineer focused on the messy, valuable part: taking models from a notebook all the way to something a business runs on 24/7.
My core work is LLM fine-tuning, computer vision, and the data pipelines behind them — plus RAG systems, chatbots and workflow automation. I've also built production voice agents for clients when a project called for it.
Currently a Data Engineer at Adalfi (Techlogix), BSc Computer Science (Data Science & ML) from GIK Institute, and IBM-certified in Machine Learning. My news-modeling research was accepted at the 31st ANLP Conference 2025 (Japan).
02 — What I do
Fine-tuning & optimizing Llama, Mistral, DeepSeek and BERT for real accuracy — measurable loss and perplexity gains on your domain data.
Production CV systems — object detection, image matching and segmentation with YOLO, CLIP, U-Net and OpenCV, deployed at scale.
Scalable ETL/ML pipelines on Spark, Airflow & Kubernetes, plus dashboards and analytics that turn raw data into decisions.
Retrieval-augmented pipelines and chatbots wired into your knowledge base — embeddings, vector databases, OpenAI & LangChain.
End-to-end automation with n8n — connecting Google Sheets, Drive, Airtable, CRMs and APIs so the busywork runs itself.
Client-proven inbound/outbound voice receptionists that qualify leads, book appointments and log to your CRM — built on Vapi, Retell & Synthflow.
03 — Selected work
CV
Hybrid computer-vision system for Dollar Industries — image–planogram matching with YOLO, CLIP, MobileNet, SSIM & pHash. 93% detection, 85% compliance, 70% faster audits.
LLM
Fine-tuned Llama 3, Mistral and DeepSeek R1 on medical reasoning data — 30% loss reduction (1.2 loss, 6.4 perplexity) for more reliable clinical responses.
Live
A private AI coordinator for funeral homes — multilingual voice + text intake, auto-organized Google Drive and server-side PDF generation. Live in production.
Live
A Shopify app that mirrors products, inventory and launch-day sales into an Airtable launch tracker within seconds — built with React Router, GraphQL Admin API and webhooks.
ANLP ’25
Fine-tuned BERT & RoBERTa on a 10k-sample news dataset (86% accuracy, +20% performance) to classify sentiment across sources — research accepted at ANLP 2025, Japan.
FYP
Mobile + web AI platform for dyslexia detection and intervention — analyzing handwriting, drawing, speech and reading with deep learning at 85% accuracy, with adaptive learning plans.
GAN
Implemented DCGANs and Conditional GANs to generate high-fidelity images from a 60k-image handwriting dataset — 30% improvement in clarity and realism.
Live
A done-for-you AI voice receptionist that answers inbound calls 24/7, qualifies leads and logs them to Google Sheets/CRM in real time, with automated SMS + email follow-up.
Live
Production AI agents that automate Airbnb hosting end to end — guest messaging, booking logic and personalized upselling — cutting response time by ~60%.
04 — Tech stack
05 — Experience
Migrated Adal Fintech's ETL & ML pipeline from legacy SSIS to a modern Spark + Airflow + Kubernetes stack, cutting pipeline latency by 65%. Built MerchandizingAI — a hybrid CV system hitting 93% detection and 85% compliance accuracy with 70% faster audits.
Built and deployed AirBot (myairbot.ai), a production automation platform for Airbnb hosting. Designed modular AI agents for real-time messaging, booking logic and personalized upselling — improving response time and booking efficiency by 60%+.
Curated a 10k-sample news dataset and fine-tuned BERT & RoBERTa to 86% accuracy (accepted at ANLP 2025, Japan). Trained an AI skin-detection system with YOLOv5 & Ollama PHI-3, processing 10k+ images at 92% accuracy.
Data Science & Machine Learning — GPA 3.61/4.00. Microsoft Learn Student Ambassador. IBM-certified in Machine Learning & AI; Python (University of Michigan).
06 — Testimonials
"Ali did a great job creating the first version of my MVP. He was responsive, professional, and able to turn my ideas into a working product."
"Muhammad did a fantastic job. He took feedback well, considered cost and speed, and delivered a well-documented solution. I'll definitely work with him again."
"His performance is extraordinary — detail-oriented, always learning the latest tech, and building AI applications that drive client satisfaction. I recommend him to anyone."
07 — Contact
I usually reply within a few hours. Tell me what you want to build and you'll get a concrete plan and timeline.
ali408mehmood@gmail.com