Sudheer B.

I build automation systems that keep running when nobody is watching them. Over five years my automation work has generated $7,700+ in annual cost savings, removed 3+ headcount of manual effort, and cut multi-hour recurring tasks down to minutes. Most automation breaks in the same three places: duplicate records, silent failures, and edge cases nobody tested. I design for those first - deduplication logic, retries, validation that flags exactly which field failed, and human review steps where the cost of being wrong is high. What I build • End-to-end workflow automation - lead capture through CRM, qualification, document generation, and follow-up • WhatsApp and messaging automation - conditional routing, notifications, scheduled reminders, owner dashboards • AI agents and multi-agent systems - LangChain, RAG pipelines, AWS Bedrock, Gemini, Google ADK • Document and data pipelines - PDF extraction, structured validation, low-confidence flagging, Excel and report generation • API integrations - REST APIs, webhooks, OAuth, and moving data cleanly between systems that were never designed to talk Recent work A WhatsApp review funnel that routes customers by rating - advocates to a public review page, unhappy customers to a private feedback form with instant owner alert. Built the routing logic, the rating page, and the real-time dashboard. A GST reconciliation tool for Indian SMBs: PDF invoice OCR, fuzzy column mapping across inconsistent vendor formats, and colour-coded Excel reports showing exactly which input tax credit is at risk. Test-covered, with low-confidence rows flagged for manual review rather than silently guessed. An AI troubleshooting bot on AWS Bedrock with a human-in-the-loop approval step before any action is taken. Stack Python, FastAPI, TypeScript, AWS (Bedrock, Lambda), Google Cloud, LangChain, ChromaDB, MCP, REST APIs, webhooks, OAuth, Streamlit, pandas. Background Five years at enterprise scale, including catalog and ML operations at Amazon across thousands of product listings - which is where I learned that automation is only useful if it handles the messy 10% of records, not the clean 90%. Fluent in English and German, native Hindi. Send me the problem and what "working" looks like, and I'll tell you honestly whether I'm the right fit before you spend anything.