At Bolt, I designed and shipped an internal quality intelligence platform that replaced fragmented Google Forms and manual coordination with one secure system for QA evaluations, dispute handling, coaching documentation, and operational reporting. I built the product end-to-end with Next.js and TypeScript, modeled workflows and data in Supabase/PostgreSQL, implemented role-based access control with row-level security for QA specialists, managers, specialists, and admins, and delivered analytics dashboards that turn raw evaluations into actionable team insights. This project combined product thinking, full-stack architecture, data modeling, security-first engineering, and internal UX design, and resulted in faster evaluation cycles, lower coordination overhead, stronger auditability, and much better visibility into performance trends and follow-up actions.

Designed and launched the VertexNova agency website as a conversion-first product, then engineered the full implementation with Next.js and TypeScript, reusable UI patterns, and responsive layouts across devices. The work combined technical SEO, Core Web Vitals optimization, and structured service messaging to improve discoverability and move visitors from first touch to inquiry through a clear, measurable funnel.

Built a predictive analytics pipeline on top of 150K ride events to identify cancellation risk and operational failure patterns before they occur. Feature engineering, statistical testing, and classification modeling produced 89% accuracy with 0.96 AUC, and the outputs were operationalized into KPI tracking for completion rate, demand volatility, and fare behavior so teams could prioritize interventions by impact.

Created an allocation and incentive-planning model using 839 hours of mobility data to map supply gaps by time window and location. The system surfaced 123 critical under-coverage intervals, estimated 397 additional driver hours for stability, and compared incentive strategies, including guarantee-based schemes with projected 345% ROI, producing execution-ready heatmaps for dispatch and growth teams.

Implemented a Tableau-based operational intelligence layer on 30K call records, with standardized KPI definitions for response time, call volume, and agent quality across shifts and channels. Parameterized views and calculated metrics exposed peak-hour bottlenecks and workload imbalance patterns, enabling leadership to rebalance staffing and improve service performance with data-backed decisions.

Developed a synthetic data platform that generated 450K realistic ride events for analytics and model testing without privacy risk. Probabilistic behavior modeling reproduced relationships among surge pricing, wait time, demand pressure, and cancellations, then exported five validated datasets with reproducible ETL steps for BI workflows and experimentation.

Orchestrated a memory-aware automation stack with n8n and integrated APIs across Gmail, Calendar, and task systems to execute multi-step operational workflows. Context persistence was implemented through Supabase pgvector memory layers and search enrichment, allowing agents to retain conversation state, make better decisions across runs, and reduce manual coordination overhead.

Engineered a hardware-software prototype of a vertical-axis wind turbine optimized for turbulent urban airflow, combining 3D mechanical design, embedded sensing, and remote monitoring. LoRa telemetry pipelines captured real-time performance signals for iterative testing, turning the build into a measurable R&D platform for energy-capture optimization in constrained environments.

Built an AI-assisted decision platform that maps student profiles to education and career pathways through structured recommendation logic and clear product UX. The system delivers personalized VET and academic track guidance, translating complex option spaces into practical next steps, and was recognized with a Startup Intern Tbilisi win for product value and execution.