Usman A.

Machine Learning Engineer. I've been building ML and AI systems in Python for six years, from search ranking and recommendation models to LLM applications running in production today. I worked on Upwork's own internal ML team and at Alibaba/Daraz, on machine learning algorithms serving millions of users. These days, most of my recent work is in the FINANCE TECH and LEGAL TECH: Financial Intelligence & Analytics, Credit Intelligence, SEC Filings Automation, financial data extraction, and legal/IP platforms. In my pipelines, code does the extraction and validation; the LLM only writes on top of verified data. 🏆 𝐓𝐨𝐩 𝐑𝐚𝐭𝐞𝐝 𝐏𝐥𝐮𝐬 | 𝟏𝟎𝟎% 𝐉𝐨𝐛 𝐒𝐮𝐜𝐜𝐞𝐬𝐬 ⏰ 𝟳𝟬𝟬𝟬+ 𝐇𝐨𝐮𝐫𝐬 | 𝟒𝟐 𝐉𝐨𝐛𝐬 | $𝟒𝟎𝟎𝐊+ 𝐄𝐚𝐫𝐧𝐞𝐝 WHAT I BUILD: ============ ✅SEC Filings Pipeline: I built a system for a finance research firm that downloads new filings from SEC EDGAR every day, parses ABS and BDC filings, validates the extracted data, and stores it in PostgreSQL. Dashboards update automatically, and the LLM generates client reports from the verified data. It has been running in production for over two years. ✅Trust Filings OCR Pipeline: This system processes trust filings that only exist as scanned images. It reads them using Tesseract and AWS Textract, parses and validates the data, calculates KPIs, and delivers AI-written reports on a schedule using Airflow and Docker. This is an ongoing Upwork contract. ✅Python Backend and Data Science: I manage a client's production backend and data science workloads end to end in a long-term engagement. ✅RAG Study Assistant: I built a chat assistant that answers questions from licensed real-estate course material and stays strictly within it. Along the way I solved the problems that only appear with real users, like retrieval confusing two similar rule sets and follow-up questions losing context. A live demo is available. ✅Tender Requirements Extraction: This system reads government tender documents and extracts a structured list of requirements for a procurement SaaS platform. ✅Trademark Case-Law Engine: I built retrieval and similarity search over EUIPO case law for an AI trademark platform, so lawyers can find similar past cases quickly. ✅ Earlier ML work includes medical forms generated from doctor-patient conversations, credit default prediction, anomaly and spam detection, and vehicle detection. I'm a Kaggle Competitions Expert, ranked in the top 5% worldwide in NLP competitions. The stack I work in daily is Python, FastAPI, TypeScript, PostgreSQL, LangChain, LangGraph, OpenAI, Anthropic API, Pinecone, Hugging Face, Tesseract, AWS Textract, Airflow, Prefect, Docker, n8n, Make and AWS. For ML work I use PyTorch, TensorFlow and Scikit-Learn. Feel free to message me anytime to discuss your project. I'm happy to do a quick call first if it helps clarify scope before we start.