Sandeep B.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'm a Data and AI engineer. On most projects I lead the architecture and also build it: the data platform, the pipelines, the AI layer, security, and the handover to the client's own team. How I work: 1. Measure before promising anything. I capture a baseline first, so every improvement I report is a real number. 2. Prove it small before committing the budget. A proof of concept with clear pass criteria comes before the main build. 3. Keep the AI on a short leash. Numbers come from code, not from the model. For anything legal, financial or customer-facing, the AI drafts and a person approves. 4. Ground answers in the client's own documents, with sources, instead of trusting what the model remembers. 5. Build it so the client can run it without me. Runbooks, training, and a test that their team can fix a live failure on their own. Please look at my portfolio to see what kind of work I do. I'm most useful when the problem is messy data spread across many systems, or when a business needs to ask questions of its own documents and records and get reliable answers. Recent work: - Fabric migration for a global production group: moved a 15-year-old SAP BW reporting estate to Microsoft Fabric. Queries got 86 to 94 percent faster, all external audit findings on data governance were closed, and the client's team ran it themselves within 8 weeks of go-live. - AWS to Azure migration for a trade data platform: 86 country datasets, about 800 GB, 1.5 to 2 billion records. Main queries dropped from 15-40 seconds to under 3, with zero downtime at cutover and per-subscriber data isolation. - Real-time analytics for a large delivery platform: Kafka into Fabric Eventhouse, replacing a pipeline that was 30 to 40 minutes behind. Dashboards now update in seconds across four live use cases. - Claims processing for a health insurer: OCR with several engines, a multi-agent extraction flow, and human review for low-confidence fields. About 150,000 pages a month, claim rejections down 68 percent. - HR analytics with an AI assistant: 10 dashboards with chat-based questions on the data, attrition forecasting and prediction, and row-level security so each HR partner sees only their own people. - Predictive quality for fibre manufacturing: ensemble models on sensor data with SHAP explanations. 87 percent accuracy and 24 percent fewer defects. - Customer assistant for an online store: answers from the product catalogue with citations, checks live order status, and replies on WhatsApp and Instagram. I've also worked on a multi-agent fashion trend research tool, a QBR report automation platform for 300+ account managers, an AI-assisted GST platform for a state tax department, and voice bot designs for a fintech company and a state helpline. Depending on the problem, my work usually involves: → Data pipelines and Medallion lakehouse design → Real-time ingestion (Kafka, Eventstream, KQL) → RAG and hybrid search over documents → Multi-agent systems (LangGraph, Semantic Kernel, MCP) → OCR and document extraction → Natural language to SQL/DAX over business data → Forecasting and classification models, with explainability → Power BI semantic models and row-level security → Evaluation sets and monitoring for AI output → Cloud migration, landing zones, CI/CD, infrastructure as code Day-to-day: Python, PySpark, SQL, KQL, FastAPI/Flask, LangChain, LangGraph, Azure OpenAI, Azure AI Foundry, Azure AI Search, Microsoft Fabric, Power BI, Azure ML, Docker, Kubernetes, Terraform, GitHub Actions, Azure DevOps, RAG, Agentic Development If you already have a system that is slow, unreliable or hard to trust, I can help find what is holding it back. If you're starting from an idea, I can help work out the simplest version worth building first, test it quickly, and grow it from there.