At Futwork, I built Echo, a sentence-level TTS audio cache for Pipecat voice pipelines. Shipped as echospeaks on PyPI, it cut TTS cost by about 40% and is now being productised as SaaS.
I also built a lead-dialing pipeline for voice agents that handled 100K+ leads a day, with per-channel rate control. I architected a MongoDB-to-BigQuery pipeline syncing 18GB daily and cut monthly BigQuery costs by 77% through query tuning.
On my headstart project, I built a real-time TTS inference server using Kokoro ONNX, FastAPI, and WebSockets. Streaming by sentence and tuning concurrent inference against measurements reduced time-to-first-audio and raised delivered throughput.
