Attila – Python, OpenAI, LangChain

An AI-powered customer-experience agent platform focused on technically complex products, initially built to solve retrieval and support challenges for auto parts and accessories.

• Architected a multi-tenant AI conversation platform across web chat, voice, and WhatsApp for EU e-commerce retailers.
• Built product retrieval over 250K SKUs using Postgres and pgvector.
• Implemented durable agent execution with Temporal for long-running tool calls.
• Designed human-in-the-loop escalation with full context handoff to support staff.
• Delivered a Dutch-language agent resolving 73% of conversations end-to-end.
• Delivered a 7-language agent that cut support costs by 40% for a Romanian retailer.
• Integrated Shopify, WooCommerce, and CCV Shop with continuous product-sync ETL.
• Built voice and WhatsApp channels on Twilio with regulatory provisioning.
• Cut inference costs via model routing, prompt caching, and usage tracking.
• Ran production infrastructure on Hetzner, Dokploy, Supabase, Redis, and Cloudflare.

A quantitative sports-betting platform for MLB and NHL markets that ingested live and historical game, team, and player data alongside sportsbook and prediction-market pricing, ran probabilistic models to estimate true outcome probabilities, and surfaced positive-edge opportunities by comparing model output against market prices — with built-in backtesting against historical odds and results.

• Built the data infrastructure and production environment from the ground up.
• Developed probabilistic prediction models for MLB and NHL game outcomes.
• Built ingestion pipelines for live and historical game, team, and player statistics.
• Collected and tracked sportsbook and prediction-market odds over time.
• Implemented expected value calculations to identify positive-edge opportunities.
• Built a backtesting framework for evaluating strategies against historical odds and outcomes.
• Improved model accuracy through feature engineering and joint probability modeling.
• Designed aggregation layers for derived player and team performance metrics.
• Orchestrated data collection and model workflows with Temporal.
• Built the React frontend surfacing model probabilities and market divergence signals.

A sustainability and supply chain risk platform for large consumer brands and their supplier networks, turning structured supplier assessments (emissions, energy, water, waste) into scored, benchmarked metrics exposed via dashboards and public APIs for ESG reporting and supplier risk management.

• Architected the data pipeline generating structured assessments and quantitative metrics from supplier responses.
• Designed and built the data warehouse supporting large-scale ingestion of assessment, emissions, and supply chain data.
• Built scoring and aggregation logic for greenhouse gas emissions and other environmental indicators.
• Designed public REST APIs exposing assessment data to enterprise customers for their reporting pipelines.
• Built and maintained a shared internal library for survey data models and business logic across services.
• Scaled the platform to handle growing assessment volume across large supplier networks.
• Implemented search and indexing over assessment data with Elasticsearch.
• Containerized services with Docker and operated the platform on AWS.

A large online food ordering and delivery marketplace connecting consumers with local restaurants (Grubhub). The work centered on the data platform behind restaurant listings — normalizing, deduplicating, and analyzing high-volume listing data from many upstream sources to keep the marketplace accurate and current.

• Built ingestion pipelines collecting restaurant listing data from numerous upstream sources at high volume.
• Processed and analyzed large datasets with PySpark on Hadoop, writing distributed MapReduce-style jobs.
• Ran distributed processing workloads on AWS EMR clusters.
• Normalized and deduplicated listing records from heterogeneous sources in inconsistent formats.
• Designed and maintained relational schemas for listing storage.
• Built Flask services and internal APIs for accessing and managing listing data.
• Contributed to deployment and infrastructure automation for data services.