Jakub Kulik

Led engineering for a UK healthcare platform and an offline commercial Pro license video-transcription/premeir extension. Built multi-instance dispatch and a load-balanced SIP server choice with clear calling-hour rules (do-not-call/calling-hours) and signed webhooks. Designed and implemented a local TypeScript panel over a Python service, with extensive automated testing, including caption segmentation using a dynamic programming approach (reading speed, punctuation, pauses, video cuts, Polish grammar) and a re-writer step validated by JSON schema checks. Built a voice-bot retrieval path (RAG with embedding-based search and a guarded output layer) that prevents unsafe or incorrect tool calls and closes safely when classifiers fail. Authored robust evaluation that runs scripted patient transcripts through production tools against the real model, and addressed root-caused production incidents with precise fixes and regression tests across billing alerts and outbound AI call pipelines. C

Worked on a cloud PBX for hundreds of UK GP surgeries, supporting NHS England advanced telephony requirements. Owned and contributed to the open-source SIP stack features across routing, queuing, recording, billing, and AI voice agents. Introduced production OpenAI function calling and associated tool usage, then led the bot module releases (prompt design, structured outputs, billing). Handled sensitive patient data under NHS assurance requirements, including EMIS and SystemOne integrations, digest authentication, access control, audit trails, and fail-closed security defaults. Built call control covering attended/blind/three-way SIP transfers, codec negotiation, DTMF handling, interception, and timeouts; fixed an RFC 3891 refer/replaces bug that could leave transfers unanswered. Built a test suite from scratch (around 1,150 tests across 99 spec files) and created a customer-facing reporting dashboard app with call volume forecasting, NHS reporting, queue statistics, shareable and temp