Omar S.Private profile: Upwork shows its earnings only to signed-in members, so we check it every few weeks, not daily. MRR here is a rough estimate (Upwork's six-month earnings ÷ months active). Last read Nov 5, 2026.

I build AI systems that convert documents into structured data, retrieve information, and answer questions with source references. I am a machine learning engineer with 7+ years of experience across retail, fintech, and eKYC. My work covers models, evaluation, application APIs, and cloud deployment. Selected results - Built an LLM call-transcription system that processed 130,000+ calls and reduced after-call work by 30%. - Created image-generation and background-removal tools that saved the design team more than 10 hours per week. - Delivered Xendity’s first internal OCR solution with object detection and eliminated third-party OCR costs. - Built an internal LLM platform with audit records, document parsing, and internal tool integrations. - Migrated recommendation-system data from MySQL to Bigtable and developed model APIs and A/B tests. Services - Document AI: OCR, structured extraction, schema validation, confidence checks, and human review workflows. - RAG: document ingestion, hybrid retrieval, source references, and evaluation of answers and retrieval results. - LLM evaluation: reference datasets, scorer reviews, regression checks, and statistical comparisons between model or prompt versions. - MCP and integrations: Python and Rust tools that connect AI assistants to APIs, documents, and business systems. - Local LLMs: MLX/oMLX setup, model comparisons, and performance diagnosis on Apple Silicon. Project work My project work includes four tools for document systems and evaluation: - evalkit: extraction and RAG evaluation with checks for regressions in individual fields. - docjson: document extraction with schema validation, confidence checks, and a review queue. - ragcite: BM25 and dense retrieval with source verification and refusal controls. - mcp-docs-server: Python MCP tools for eCFR, DailyMed, and local document pipelines. Open-source work and contributions - Parsnip: a Rust memory graph for AI assistants with MCP integration and text search. - atlassian-cli: Rust tools for Jira, Confluence, Bitbucket, and Jira Service Management. - gemini-mcp-rust and reddit-mcp-server: MCP integrations with external APIs. - llm-benchmark: model evaluation, long-context tests, and checks for invalid or truncated output. - bengali-ocr-finetune: MLX and LoRA experiments with documented evaluation results and limits. - App-Store-Connect-CLI: merged contributions for validation, localization, and pricing features. Application projects Application projects include BankStatementReader, which combines Python/FastAPI OCR, React, and Cloudflare infrastructure. MyDay uses Workers AI, Vectorize, and D1 for note retrieval and chat. Technologies and certifications Core technologies include Python, Rust, TypeScript, Go, PyTorch, TensorFlow, GCP, AWS, and Cloudflare. I hold AWS and GCP certifications. Send a sample document, an example query, or a description of the system you need. Include the required result and deployment environment.