Deepanshu Goel

— Built FlowOS end to end for an industrial instrumentation manufacturer. Its sales team now runs quoting on it, from a customer email in a shared Microsoft 365 inbox to a priced quotation PDF they can revise and reissue.

— Designed its LLM pipeline, which finds the real customer behind a forward, OCRs attachments and scanned PDFs, and classifies each enquiry with the lines it relied on. It replaced a proof of concept whose character cap lost 24–78% of long enquiries.

— Shipped multi-agent systems for radiology report interpretation and for construction scheduling grounded in project data.

— Own engagements from discovery through production — workshops, iteration, and live adoption.

— Built reusable multi-agent orchestration patterns so later customers start further up the curve.

— Build Rinne with Siddharth Manjul: an open-source Rust CLI and macOS app that plans a goal as a DAG and drives Claude Code, Codex, Grok, and OpenCode as workers until the checks pass.

— Shipped the TUI's live agents view and command palette, the curl installer, and rinne learn, which indexes a codebase into a code graph and explains it for engineers and for product people.

— Released through v0.1.9 on crates.io. Workers share state on disk, and API keys stay in the OS keychain.

— Architected and shipped the flagship EdTech platform for professional training programs (healthcare workers, navy personnel, World Bank–backed cocoa farming education).

— Added LLM conversational features, automated content analysis, and performance summarization — engagement up ~25%, learner retention up ~15%.

— Shipped reusable JavaScript SDKs and cloud-native CI/CD; client integration effort down ~50%, deploy time down ~40%.

— Built real-time analytics so training performance data could drive continuous product decisions.