Mike Polinowski

Research Assistant & Applied Astrophysics / Plasma Physics

2005 - 2010

University of Cologne, Germany

Mass spectrometry & Terahertz spectroscopy for iron-compound detection in low-pressure, high-energy plasmas. Teaching position — inorganic-chemistry laboratory courses.

M.Sc. & Applied Astrophysics / Plasma Physics

2002 - 2004

University of Cologne, Germany

Mass spectrometry & Terahertz spectroscopy for iron-compound detection in low-pressure, high-energy plasmas. Teaching position – inorganic-chemistry lab courses.

B.Sc. - Atmospheric Science & Biological Chemistry

2000 - 2002

University of Wuppertal, Germany

Spectroscopic methods for atmospheric analysis (Physical Chemistry). Teaching position — instrumental analysis & inorganic chemistry.

Chief Technology Officer (AI & Infrastructure Lead)

2016 – Present

INSTAR Deutschland GmbH (Waletech)

Hongkong S.A.R. / Shenzhen, China

Video-surveillance and smart-automation manufacturer. Own the full technical stack — from embedded firmware through cloud infrastructure to production AI/ML — in a small, fast R&D team.

• AI Agents & LLM Integration (2026)
• Built a Model Context Protocol (FastMCP) server over the camera control API, exposing six domain-level tools that let any MCP-compatible LLM agent drive the hardware — without touching raw protocol, and with state-confidence tracking so the agent reports only what it confirmed.
• AI / MLOps & Inference Optimization (2020 – Present)
• Built an end-to-end machine-learning pipeline for a CPU-only AI inference platform — from data ingestion and training / fine-tuning through a version-controlled model registry to production model serving — monitored for performance and reliability.
• Designed a distributed inference pipeline (Ray actor pool, Redis job queues, cross-server load balancing) that serves real-time object detection (YOLO), ANPR / OCR, and a fine-tuned Vision-Language Model; battle-tested at 10,000+ detection tasks per minute.
• Drove a significant cost-and-reliability result: migrated a multi-GPU TensorFlow/PyTorch stack to a two-CPU-server architecture, maintaining or improving accuracy while cutting compute cost by ~10× and eliminating crashes, memory leaks, and peak-load stalls.
• Fine-tuned and quantized VLM and CV models for edge and cloud (OpenVINO, LiteRT / TFLite, int8 / float16) to hit latency and accuracy targets on hardware without a GPU.
• Shipped customer-facing computer-vision prediction APIs and an Elasticsearch-backed search layer so alarm, plate, and face events become text-searchable.
• DevOps / DevSecOps / Platform (2018 – Present)
• Run a 70+ bare-metal and VPS fleet across multiple datacenters, orchestrated on HashiCorp Nomad + Consul: self-healing, auto-scaling, canary-first deployments with automatic rollback and least-privilege, per-namespace ACLs (mTLS server↔client).
• Standardized CI/CD on self-hosted GitLab (source → Docker → registry → Nomad), replacing manual SSH deploys as the single path to production.
• Own observability and reliability end to end — Zabbix, Grafana, Kibana / Filebeat with per-service dashboards, alerting, and runbooks — so the team is paged on degradation, not on a customer ticket.
• Owned DevSecOps across every customer-facing service: TLS 1.3 / HSTS, network segmentation, access control, CVE response, backup/restore and disaster recovery.
• IoT & Product / Full-Stack (2012 – Present)
• Extended the product into smart-home automation by designing MQTT (MQTTv5) and webhook interfaces in firmware — making INSTAR the only surveillance-camera maker exposing full control over MQTT — and remain the first-escalation contact for those integrations.
• Earlier built the company's customer knowledge base, community forum, and SLA-tracked ticketing from zero to production (React / Gatsby, Elasticsearch, Discourse, osTicket, Go/Fiber) and integrated them across all in-house products.

General Manager / Platform Lead

2012 - 2015

INSTAR Deutschland GmbH

Guangzhou, China

Managed a 5+ person team and 5 concurrent projects under tight deadlines while the company scaled its online presence and customer-support tooling.

• Full Stack WebDev / Product QA
• Built the company's first structured digital ecosystem from zero — knowledge base, community forum, and SLA-tracked support platform — while integrating each into all in-house products.
• Delivered concurrent web and support projects under commercial deadlines as a cross-functional lead; designed internal training programs for customer-service and production teams.
• Coordinated new-product development and QA with the production line, acting as the liaison between engineering, manufacturing, and sales.

Support / QA Technician

2011 - 2012

INSTAR Deutschland GmbH

Guangzhou, China

First technical hire on the China side; covered the full support-and-QA cycle from bench to production line.

• Full Stack WebDev / Product QA
• First technical hire on the China side; delivered Tier-1 / Tier-2 support and on-site quality assurance for network and IoT hardware during initial scale-up.
• Authored customer-facing documentation and training; enabled integration of INSTAR hardware into third-party software and security platforms; performed firmware QA / burn-in and fed failure modes back to design.