Muhammad Usman B.

I build AI and IoT systems that survive contact with production - in hospitals, on industrial sites, and inside enterprise data platforms. Clients hire me when a prototype has to become something clinicians, operators, and auditors can actually rely on. I do the architecture and the implementation. There is no handoff between the person who designs the system and the person who has to make it work at 2am. CURRENT WORK • Sental IQ - hospital-deployed healthcare IoT and vitals platform Deployed at Gold Coast University Hospital in Australia. I designed the full path from device to clinical record: an on-premises edge gateway (Docker, Mosquitto MQTT, Telegraf, HTTPS and TCP adapters) collecting from mmWave fall-detection sensors, time-of-flight people counters, air-quality units and clinical wearables; a real-time vitals pipeline moving blood pressure and biometric data through webhook push instead of polling; and an AWS backend with FHIR R4 as the clinical system of record and InfluxDB as the derived time-series layer. The models were never the hard part. The hard part was device classification across LoRaWAN, PoE and vendor-cloud connectivity, distinguishing patients from staff in top-down ToF heatmaps, and making every reading traceable enough for a clinician to trust. • Medical Equipment Finder - AI discovery and decision platform Build for a US medical equipment refurbisher: an AI system that helps buyers find and evaluate equipment, with retrieval grounded in real inventory and specifications rather than model guesswork. SELECTED EXPERIENCE • Built an industrial IoT telemetry pipeline ingesting 70,000 tags from Azure IoT Hub through Event Grid, Stream Analytics and Data Lake for Ashghal Authority Qatar (a national infrastructure authority) - 20% improvement in ingestion efficiency. • Developed a generative AI platform for ad-hoc data analysis in natural language. Ran a POC across multiple LLMs (Olama, Hugging Face, Kimi, DeepSeek, etc) before selecting OpenAI, improving query understanding by 60%. • Delivered 88%-accurate forecasting on IoT and business data using AutoML and scikit-learn, served through a Python/Flask and Azure architecture. • Built an AI medical record summarizer end to end for a US healthcare client using Agentic RAG. • Advised a client team from GMC Motors UAE on Databricks AI agent architecture. • Founded myNtropy, a B2B SaaS that turns LinkedIn networks into structured intelligence: FastAPI on Azure, React, Azure SQL, OAuth 2.0, Stripe, GPT-based enrichment, CI/CD on GitHub Actions. Built solo, end to end. STACK • Machine Learning - time-series forecasting, classification, clustering, anomaly detection, feature engineering, AutoML, model evaluation and hyperparameter tuning (scikit-learn, XGBoost, pandas, NumPy, Azure ML) • Deep Learning - CNNs for image and sensor-frame analysis, RNNs and LSTMs for sequence and telemetry data, transformer architectures and attention mechanisms • LLM Systems - RAG pipelines, agent architectures, tool use and MCP, prompt and context engineering, evaluation harnesses and hallucination guardrails (OpenAI, Claude, vector databases) • Azure - IoT Hub, IoT Edge, Data Factory, Machine Learning, Data Lake Gen2, Functions • AWS - IoT Core, Lambda, S3 • Data & Streaming - InfluxDB, Neo4j, Grafana, Databricks, Event Grid, Stream Analytics, MQTT, Kepware OPC, FHIR R4, SQL • Engineering - Python, FastAPI, Flask, Node.js, React, Supabase, Docker, GitHub Actions, Next,js HOW I WORK I start with the workflow and the cost of failure, not with a model or a vendor. I ask what evidence the system has to produce and who has to trust it, then design backwards from there. When an AI answer can be confidently wrong, I build the checks that catch it before a user sees it. I am comfortable owning the whole path from discovery to production, or embedding with an existing team to solve the specific architecture, integration or reliability problem that's blocking them. Every rated contract I have delivered on Upwork is 5.0. What clients flag most often is communication, reliability and accountability. One put it simply: "delivered high quality work and demonstrated strong accountability." I hold a BS in Data Science from FAST-NUCES, Pakistan, and run DATXOC, my own engineering practice. If you have an AI or IoT system that has to work in the real world, tell me what breaks today and I'll tell you honestly whether I'm the right person for it.