PeriodOct 2019 – Jul 2021
LocationNew York, NY / Odesa, Ukraine
Impact50% faster production
Websitesembly.ai/
Tech Stack
KafkaNLPFastAPIPythonMicroservicesDocker
About the Company
Sembly AI is an enterprise SaaS platform that transcribes and analyzes meeting audio, using NLP to extract action items, summaries, and decisions.
My Journey There
I worked on the high-load ingestion pipelines, designing a Kafka-based broker infrastructure to parse parallel audio streams.
I split the old monolithic API into smaller microservices using Docker and FastAPI, achieving a 50% increase in product release speed.
Optimized the GPU inference scheduling logic for NLP models to reduce token latency by 25%.
Key Deliverables & Architectures
- [01]Established end-to-end automated training and validation workflows, cutting model time-to-production by 50% and standardizing deployment cycles.
- [02]Architected a high-throughput audio ingestion system using Apache Kafka, enabling real-time processing and driving a 30% improvement in system responsiveness.
- [03]Overhauled NLP inference pipelines, achieving a 25% speed boost and a 15% increase in model performance.
- [04]Decoupled the monolithic AI platform into fault-tolerant microservices, enhancing system resilience and slashing deployment latency by 40%.