Andrzej – Python, LLM, AI agent development

An enterprise AI agent development platform built for a global manufacturing corporation, enabling business units to develop and deploy LLM-powered and other AI agents. The platform provides a shared SDK, agent orchestration engine, evaluation tools, and low-code development capabilities to support AI solutions tailored to different business needs.

• Developed the core SDK and agent orchestration engine as one of the founding engineers, becoming the project's second-largest contributor;
• Collaborated directly with business units to gather requirements and adapt AI solutions to their operational needs;
• Contributed to the agent evaluation framework, CLI, low-code agent builder, and developer tooling across multiple engineering teams;
• Developed agentic prototypes for manufacturing production forecasting and real-time visitor guidance;
• Trained a gradient-boosted purchase-intent model for visitor guidance, achieving approximately 0.95 AUC and 80× lift among top-ranked leads.

An AI-powered in-flight assistant developed for a global aviation technology provider and designed for aircraft seat-back entertainment systems. Built on an enterprise agent platform, the solution combines a conversational avatar, voice interaction, and layover planning to help passengers manage travel disruptions, arrange accommodations and transportation, and plan local activities.

• Developed an AI-powered in-flight assistant using an enterprise agent platform, integrating conversational AI, an interactive avatar, voice capabilities, and layover planning;
• Delivered the solution for integration with aircraft seat-back entertainment systems;
• Demonstrated the product to representatives of major international airlines at Aircraft Interiors Expo.

An LLM-powered family wellness assistant developed for a global technology company, designed to help families coordinate shared goals, schedules, and personalized recommendations.

• Led the development of the multi-agent conversational AI system from scratch as the project's primary developer and largest contributor;
• Implemented agent coordination capabilities for managing family goals, scheduling, and personalized recommendations;
• Advanced the product through internal dogfooding and trusted-tester programs with real families;
• Developed shared agent libraries and reusable service and agent templates adopted across the engineering team;
• Collaborated directly with Anthropic engineers during a strategic partnership to support AI development.

A fully offline neural text-to-speech (TTS) system developed for in-vehicle voice applications at major automotive manufacturers. The solution enables natural-sounding speech synthesis directly on embedded devices without requiring an internet connection, while maintaining a compact model size and low audio-generation latency.

• Developed the core of an offline neural text-to-speech engine deployed in production vehicles;
• Served as technical lead of the on-device TTS engineering team, guiding technical development and implementation;
• Optimized English neural voice models to fit within 150 MB while achieving approximately 2× real-time synthesis speed and under 200 ms latency to first audio on Pixel 4 XL-class hardware.