Case studies in industrial data acquisition, production machine learning, AI services and the infrastructure required to operate them.
Built read-only data acquisition pipelines for industrial equipment, moving PLC telemetry from edge collectors into ClickHouse for monitoring, analytics and downstream ML.
Built ML pipelines on real manufacturing telemetry, from time-windowed PLC datasets and defect modeling to online inference and shadow-mode validation on production equipment.
Mapped thousands of low-level PLC registers to real machine nodes and process semantics using historical telemetry, electrical documentation and field validation.
Deployed an industrial knowledge assistant combining RAGFlow, local embedding/reranking models and LLM inference to search and reason over technical machine documentation.
Built an internal OCR service for structured document extraction with validation, error handling and asynchronous integration into 1C workflows.
Built and deployed a voice assistant pipeline for meeting transcription and AI processing on NVIDIA Jetson, packaged with Docker and exposed through a secured web service.