pavelmoroz

Built and launched automation products, AI agents, and supporting infrastructure for a fast-moving US healthcare product, supporting patient communication, lead management, internal workflows, and operational efficiency: • Translated business and operational needs into product initiatives, combining discovery, requirements, user journey design, technical alignment, delivery sequencing, launch criteria, production monitoring, and post-release iteration. • Built and launched AI-powered communication platform for channels across text and voice, integrated with internal systems, patient workflows, lead management, and support operations. • Shaped production architecture and agent behavior using LangGraph, LangChain-style pipelines, LLM tool calling, structured outputs, API/tool contracts, deterministic backend rules, prompt/version management, and model/prompt benchmarking across dozens of configurations. • Built product quality and evaluation practices for AI agents and tools, running thousands of scenario-level simulations across hundreds of curated cases, with custom test frameworks, golden conversations, regression checks, tool-call correctness validation, output quality review, and incident-to-test loops. • Established observability and release practices with traces, prompt/model/tool versioning, feature flags, controlled rollout, latency/cost monitoring, and production issue analysis.| • Embedded healthcare-specific constraints into product logic, including PHI-aware design, safe escalation, human handoff, uncertainty handling, and deterministic enforcement of business rules in backend systems.

Led product and technology development for large-scale text and voice assistants across telecom service scenarios, owning the product platform roadmap, automation growth, quality metrics, and cross-functional delivery: • Translated enterprise customer requirements into product scope, roadmap priorities, delivery plans, and investment/resource estimates, helping prioritize high-impact automation initiatives and align product development with business outcomes. • Built and led a 50-person cross-functional product team, shifting development from a resource-based model to a dedicated product structure with clearer ownership, delivery routines, and KPI-driven priorities. • Defined and executed the product roadmap for conversational automation, including ML-based classification, sentiment analysis, satisfaction prediction, user segmentation, and LLM adoption. • Established product quality metrics and analytics frameworks across automation, Voice of Customer, GCR, FCR, classification quality, and A/B testing. • Led the first commercial LLM integration for conversational AI within MTS Group, increasing automation by 10% while maintaining user satisfaction. • Improved assistant comprehension by 7% in voice and 24% in text through ML-driven features, classifier optimization, and structured evaluation. • Oversaw the training of 667+ model versions and deployment of 20+ custom production models for classification, sentiment analysis, and satisfaction prediction. • Built end-to-end data-driven improvement processes, including human and AI annotation workflows for 500,000+ user phrases, BI reporting, metric analysis, and production impact evaluation.

Progressed across product analytics and platform leadership roles, working on banking assistant automation, NLU optimization, platform migration, release validation, personalization, and advanced assistant UI capabilities. • Led the development and evolution of a conversational AI platform for banking assistants, focusing on automation, personalization, NLU quality, scalable assistant architecture, and stable release processes. • Managed platform migration, ensuring backward compatibility, ML model support, scenario logic consistency, and release reliability across banking assistant scenarios. • Designed personalization and integration logic for banking assistants, including user segmentation, adaptive automation levels, buttons, tooltips, widgets, and JSON-based integration requests. • Created test datasets and evaluation processes to measure pre-release model quality using F1 score, accuracy, recall, and scenario-level performance metrics. • Automated release validation through pre-release testing, error analysis, and release guidelines for assistant scenarios and ML models, improving deployment reliability and reducing production issues. • Launched AI-driven tooltips across banking channels, used in 5-10% of all sessions and contributing to a 3-5% automation increase. • Developed predictive conversation flows using user data, increasing automation by 20% in key scenarios; contributed to Alfa-Bank ranking #1 in chatbot development among Russian banks by Markswebb in 2022.