Muhammad
Saif Ur
Rehman
I build intelligent systems - RAG pipelines, multi-agent orchestration, voice AI, and no-code automation. End-to-end. Actually ships.
About Me
I'm an AI Engineer and BS AI graduate from COMSATS University Islamabad, passionate about building systems that work in the real world, not just in notebooks.
My core expertise spans end-to-end AI pipelines: fine-tuning open-source LLMs, building RAG-based production systems on AWS, and orchestrating multi-agent workflows with CrewAI.
I'm currently deep into voice AI agents (Vapi & Retell) and n8n automation, integrating AI into real business workflows. The goal: make AI actually useful for people who aren't engineers.
When I'm not coding, you'll find me studying how frontier AI systems are architected and figuring out how to build the next generation of them. Outside of AI, I'm into gaming, fitness, and cycling.
What I Build
End-to-end retrieval-augmented generation pipelines covering document ingestion, vector indexing, LLM integration, and cloud deployment. Built to be accurate and production-ready.
Autonomous agent pipelines using CrewAI and LangGraph. Research, analyst, and writer agents working in concert to automate complex multi-step cognitive tasks.
Intelligent voice agents via Vapi and Retell for customer service, booking, and lead qualification. Natural conversations with LLM reasoning behind every response.
Visual AI workflow automation with n8n for CRM integrations, email AI, document processing, and lead nurturing, all triggered by real business events.
Custom CV models for detection, segmentation, and classification using YOLOv8 and PyTorch. From custom dataset training to real-time video inference.
Adapt open-source models to your domain using LoRA/PEFT via HuggingFace. Efficient fine-tuning on custom datasets without needing a massive GPU cluster.
My Journey
Projects
Co-led development as project leader in a 2-person team. Built a health and lifestyle advisor around 7 autonomous AI agents, each owning a distinct domain: personalised workout planning, adaptive meal planning, multimodal RAG coaching, real-time vitals monitoring with anomaly alerts, live posture analysis with voice guidance, and AI-driven blog generation. Includes a progress dashboard with visual charts and gamification (badges, streaks).
Production-deployed RAG medical chatbot on AWS. Ingests medical knowledge bases into Pinecone, retrieves context per query, and augments LLM responses for accurate, grounded medical answers. Full pipeline from HuggingFace embeddings to SageMaker endpoint.
Multi-agent CrewAI system that autonomously validates business ideas. A researcher gathers market data, an analyst evaluates viability, and a writer produces a comprehensive investment report with zero human intervention.
Dual-agent CrewAI pipeline where a research agent deep-dives any topic, passing structured findings to a writing agent producing publication-quality blog posts. Clean agent handoff and task decomposition.
YOLOv8 semantic segmentation model detecting and delineating potholes in real-time video. Trained on custom dataset with production-ready inference pipeline for smart city applications.
Parameter-efficient fine-tuning of open-source LLMs using LoRA/PEFT. Adapts foundation models to domain-specific tasks efficiently without needing a massive compute budget.
Real-time gesture classifier recognizing multiple hand gesture classes from live camera input. Shows ML training, custom dataset pipelines, and live inference, applicable to HCI and accessibility.
Built a full-stack Alzheimer’s detection application as a freelance project. Trained LeNet, VGG16, MobileNetV2, ResNet, and a custom architecture on an Alzheimer’s dataset. The custom model outperformed all benchmarks and was integrated into the final product. Built Flask backend + React frontend with real-time inference.
Built an n8n + Vapi workflow that reads a Google Sheet of sales contacts, automatically calls clients with no follow-up logged, and after each call has the AI agent summarize the conversation, update the Follow-Up Summary, and mark the contact as followed up. Eliminates manual tracking overhead for sales teams.
Built an n8n automation that manages customer support entirely through Gmail. Processes incoming messages, resolves queries automatically, and for defective item cases generates a unique RMA return number sent back to the customer.
Website-embeddable RAG chatbot that retrieves from company documents stored in Pinecone and generates accurate context-aware responses to FAQs and business queries. Plug-and-play for any business website.
Recognition
"Completed a 6-month AI internship at NCAI Medical Imaging Lab, contributing to fine-tuning VLMs and Swin Transformers on medical imaging data."— NCAI · National Centre of Artificial Intelligence, COMSATS Islamabad
"Innoquest Cohort 1 Graduate — selected for intensive AI training focused on applied ML and shipping production systems beyond academic exercises."— Innovista Rawal · AI Training Program
Skills & Stack
Let's Build Together
Open to opportunities.
Looking for AI Engineering roles, freelance automation projects, and collaborations on agentic or voice AI. Based in Islamabad, open to remote worldwide.