Kirill Mizonov

Architected and scaled the internal software ecosystem and data infrastructure for a multi- location international clinic, transitioning fragmented operations into an integrated, automated, and AI-driven enterprise. Case 1: Event-Driven Lead Management & DWH Context: Clinic with 350+ KommoCRM fields suffered API overloads and lead loss from fragmented marketing sources (Tilda, Jivo, PHP). Task: Build fault-tolerant lead ingestion and a stable DWH. Action: Deployed an isolated Flask app (Heroku) with message queues to decouple ingestion from CRM. Engineered a PostgreSQL/SQLite data mirror with an incremental ETL pipeline syncing only modified records. Added an RBAC-protected web UI. Result: Zero lead loss during peak traffic, API overhead reduced by 90%, and report generation cut to <2 mins. Stable for 4+ years. Case 2: AI Speech Analytics & QA Pipeline Context: Manual QA scoring of thousands of monthly VoIP calls was subjective, slow, and unscalable. Task: Automate compliance scoring for 100% of calls across 10+ KPIs. Action: Engineered a Python LLM pipeline. Integrated OpenAI Whisper for dual- channel time-stamped transcription. Deployed a local Llama 3 instance on mid-tier GPUs (Edge AI) for cost-efficient dialog scoring. Developed a PyQt6 analytics dashboard. Result: Automated 100% of call monitoring, reducing analysis time to <1 min per call while saving thousands in cloud API fees. Case 3: Real-Time Resource Scheduling ERP Context: Multi-location clinic suffered from patient booking conflicts due to overlapping logistical constraints (beds, equipment, conditions). Task: Develop a real-time scheduling ERP module to eliminate overbooking. Action: Used AI-assisted development for front-end (Vanilla JS, Bootstrap, WebSockets) for instant data sync. Built a Flask back-end mapping CRM data against clinical constraints. Integrated RBAC and a custom payroll module. Result: Eliminated patient overbooking and drastically cut Time-to-Market via AI- assisted engineering.

Contributed to a proprietary ERP system within a team of 2,500+ engineers, adhering to production-grade clean code, microservices architecture, and strict testing. Case 1: NLP Resume Parsing Microservice Context: Recruitment module needed to parse millions of unstructured resumes (Word, PDF) from external job boards into a unified internal format. Task: Develop high-load data ingestion features within a 20+ microservice pipeline. Action: Engineered backend components using strict Python. Implemented NLP tokenization rules and managed high-velocity data indexing across PostgreSQL and ElasticSearch clusters. Result: Secured stable processing and integration of millions of applicant profiles. Case 2: High-Load SQL Query Optimization Context: Complex search queries inside the tracking module hit severe performance bottlenecks due to millions of database rows, slowing down enterprise users. Task: Refactor legacy database interactions and optimize query execution. Action: Profileed slow query executions, refactored multi-layer legacy SQL queries, and optimized indices inside PostgreSQL to prevent memory spikes under intense load. Result: Drastically accelerated query execution times and reduced database overhead. Case 3: CI/CD Pipeline & Test Coverage Context: With thousands of developers, unsafe code delivery could introduce critical regressions into the core enterprise ERP platform. Task: Ensure reliable feature integration through automated testing. Action: Authored comprehensive backend unit/integration tests for every deployment. Integrated automated test pipelines into Jenkins CI/CD and monitored health via Grafana. Result: Established a zero-regression deployment cycle for all assigned modules.

Scaled operations from a manual paper-based business model into a data-driven enterprise, expanding the network from a local operation to 300+ employees. Case 1: Stylist Performance Tracking & Data Culture Context: High reliance on 100+ individual stylists, but tracking performance manually took hours of spreadsheet management. Action: Wrote a custom VBA batch-processing script to filter raw data and instantly generate hundreds of individual efficiency charts, showing stylists their personal statistics. Result: Created a transparent, competitive ecosystem that naturally incentivized staff and increased employee retention. Case 2: Supply Chain & Inventory Optimization Context: Locations held 80kg of dye overstock due to fear of running out, leading to expired products and financial leakage. Action: Eradicated overstocking by establishing precise consumption norms, designing smart order forms, and redistributing excess materials across salons. Result: Eliminated inventory waste, secured 100% reliable data, and allowed volume- based discount negotiations with global beauty brands. Case 3: MarTech Scraper & Data-Driven Acquisition Context: Traditional marketing channels were inefficient, and allocating budgets to social media communities was a guessing game. Action: Built a Python scraper to analyze audience density and engagement rates in local groups. Automated link tracking via strict UTM tags and calculated precise metrics. Result: Automated community scoring, allowed structured A/B tests, and accurately mapped CAC and LTV to focus budget solely on profitable channels.