Salman M.

I'm a Computer Vision and AI engineer building systems that hold up on real cameras, not just test footage, object detection, tracking, OCR and inspection, from live CCTV feeds to production lines. Top Rated | 100% Job Success | 2,000+ hours delivered On the automation side, I build AI agents and n8n automations that read emails, documents and web portals, then create the quote, update the CRM or open the ticket. I work on the problems that break naive pipelines, people blocking each other in crowds, defects a few pixels wide, documents that change layout, and workflows where a wrong answer costs money. My focus is understanding those failure cases, validating the results and adding human review where needed. 🔧 WHAT I'VE BUILT RECENTLY ▸ Real-time AI Surveillance and Event Detection: Detection agents for a commercial security platform using existing CCTV camera feeds. These identify events such as intrusion, loitering, unauthorized gatherings, property damage and fire, read license plates (ANPR), and generate alerts for review. ▸ Bid Monitoring and Quote Automation: A workflow that monitors tender emails, screens opportunities against the company’s services, processes bidding documents and creates quotes in an ERP. It also checks live bids for amendments before their deadlines. Took 16 hours a week off every assistant on the bidding desk. ▸ AI Powered Retail Analytics System: Turns ordinary CCTV cameras into footfall counts, heatmaps, dwell time per area and queue tracking for shops and cafes, with custom email alerts. It helps owners plan staffing around busy hours, improve store layout and stop losing customers. ▸ Pose-Based Form & Posture Analysis: Scores body position and movement from ordinary video against configurable correctness rules. Built for coaching, physiotherapy, and workplace ergonomics. ▸ Document & Receipt Data Extraction: Pulls structured fields out of receipts, invoices, and forms without per-template rules. Low-confidence pages route to human review instead of failing silently. 👁️ COMPUTER VISION & MACHINE LEARNING - Object Detection & Tracking (YOLO, RT-DETR, Faster R-CNN, ByteTrack, BoT-SORT, DeepSORT) - Face Recognition and license plate recognition (ANPR) - Visual Inspection & Defect Detection for Manufacturing and Quality Control - Image Segmentation (SAM, SAM 2, Mask R-CNN, semantic and instance) - OCR & Intelligent Document Processing (PaddleOCR, DocTR, EasyOCR, Tesseract) - Pose Estimation & Real-Time Video Analytics (OpenCV, MediaPipe, Roboflow Supervision) - Custom deep learning model training with PyTorch and Ultralytics YOLO, including SAM-assisted dataset labeling - Vision-Language Models & Visual Search (CLIP, Claude Vision, GPT-4o, Gemini Vision) - Edge AI & Real-Time Deployment (ONNX, TensorRT, model optimization) I evaluate detection quality, missed events, tracking consistency and speed on your actual footage and hardware, not on a benchmark dataset. 🤖 AI AGENTS & WORKFLOW AUTOMATION - AI Agents & Agentic Workflows (LangGraph, LangChain, CrewAI, MCP integrations) - Workflow automation with n8n and Zapier, plus custom Python services - Email, web-portal and document processing automation - Web Scraping & Automated Data Collection - RAG Pipelines over company knowledge bases (Pinecone, ChromaDB, FAISS, pgvector) - OpenAI, Anthropic and Google APIs, plus open-source LLMs via Hugging Face - LLM Fine-tuning (LoRA, QLoRA, PEFT) and prompt optimization 🔌 INTEGRATIONS I connect outputs to the tools your team uses, ERP and CRM systems, helpdesk and ticketing tools, Slack and email, Google Sheets and Airtable, accounting systems, and anything with a REST API. Email and webhook triggers, scheduled runs, and REST or GraphQL endpoints for the systems that do not have a connector. ⚙️ DEPLOYMENT - REST APIs and services (FastAPI, containerised with Docker) - Cloud deployment on AWS (EC2, S3, Lambda, SageMaker) - Unattended operation: scheduled and event-driven runs, retries, logging, duplicate handling, alerting, and an audit trail of what the system decided and why - Internal tools and dashboards (Streamlit) - MLflow, Weights & Biases, model versioning and monitoring ✅ STACK Python, PyTorch, Ultralytics YOLO, OpenCV, MediaPipe, Roboflow Supervision, Image Processing, SAM, CLIP, PaddleOCR, n8n, Zapier, LangChain, LangGraph, OpenAI & Anthropic APIs, FastAPI, Docker, ONNX, TensorRT, AWS I benchmark before I choose. Trackers, detectors and thresholds are selected by measured results. Automations are judged by how much they handle independently and what they get wrong without flagging it. A quietly wrong workflow is worse than one that handles less and says so. I’ll tell you if an approach is wrong for your data or process before you’ve paid for it. Have footage, documents, or a workflow that keeps eating your team's time? Send me a sample. You'll get an honest answer on whether it's solvable, how I'd build it, and where it could break, before you spend anything.