Applied AI Investment Research in L/S Discretionary

𝐌𝐮𝐥𝐭𝐢-𝐌𝐨𝐝𝐞𝐥 𝐆𝐚𝐦𝐞 𝐅𝐞𝐞𝐝𝐛𝐚𝐜𝐤 𝐏𝐥𝐚𝐭𝐟𝐨𝐫𝐦

Led the design and delivery of a platform processing 100M+ player feedback across 9 game titles from In-Game sources, Steam, Discord, Reddit, YouTube, and more. Built scalable ETL pipelines to aggregate diverse data and implemented hybrid sequential/concurrent ML workflows for translation, sentiment, toxicity, actionability extraction, and PII detection, reducing total runtime from 6 hours to 2 hours.

Provided real-time, actionable dashboards for game leadership, developers, and marketing teams, enabling rapid bug fixes and data-driven feature planning. Insights from the platform directly influenced decisions, such as fixing in-game bugs that increased weekly positive sentiment by 23%, and guided marketing strategies for future content launches.

𝐃𝐢𝐬𝐭𝐫𝐢𝐛𝐮𝐭𝐞𝐝 𝐖𝐨𝐫𝐤𝐟𝐥𝐨𝐰 𝐎𝐫𝐜𝐡𝐞𝐬𝐭𝐫𝐚𝐭𝐢𝐨𝐧 𝐄𝐧𝐠𝐢𝐧𝐞

Designed and implemented a distributed, multi-node test orchestration platform for 6 game titles across 150 consoles (PS5, Xbox, Switch) and 80 PCs. Engineered parallel workflows and dynamic configurations to process 5,000+ automated tests per day with 99.5% pass rate, reducing manual testing effort and accelerating release cycles. Leveraged Temporal-like workflow orchestration, async concurrent tasks, and dynamic configuration templates to enable high-throughput, reliable multi-game operations.

𝙏𝙚𝙘𝙝 𝙃𝙞𝙜𝙝𝙡𝙞𝙜𝙝𝙩𝙨: 𝘈𝘞𝘚 𝘚𝘢𝘨𝘦𝘔𝘢𝘬𝘦𝘳, 𝘊𝘰𝘮𝘱𝘳𝘦𝘩𝘦𝘯𝘥, 𝘉𝘦𝘥𝘳𝘰𝘤𝘬, 𝘛𝘳𝘢𝘯𝘴𝘭𝘢𝘵𝘦, 𝘖𝘳𝘭𝘦𝘢𝘯𝘴, 𝘙𝘦𝘥𝘴𝘩𝘪𝘧𝘵, 𝘚3, 𝘋𝘺𝘯𝘢𝘮𝘰𝘋𝘉, 𝘖𝘱𝘦𝘯𝘚𝘦𝘢𝘳𝘤𝘩, 𝘊𝘭𝘰𝘶𝘥𝘍𝘰𝘳𝘮𝘢𝘵𝘪𝘰𝘯, 𝘔𝘶𝘥𝘉𝘭𝘢𝘻𝘰𝘳, 𝘋3.𝘫𝘴, 𝘋𝘪𝘴𝘵𝘳𝘪𝘣𝘶𝘵𝘦𝘥 𝘌𝘛𝘓, 𝘙𝘦𝘢𝘭-𝘵𝘪𝘮𝘦 𝘕𝘓𝘗/𝘔𝘓 𝘢𝘯𝘢𝘭𝘺𝘵𝘪𝘤𝘴, 𝘚𝘵𝘳𝘦𝘢𝘮𝘪𝘯𝘨, 𝘉𝘢𝘵𝘤𝘩 𝘗𝘳𝘰𝘤𝘦𝘴𝘴𝘪𝘯𝘨, .𝘕𝘌𝘛, 𝘔𝘶𝘭𝘵𝘪-𝘕𝘰𝘥𝘦 𝘖𝘳𝘤𝘩𝘦𝘴𝘵𝘳𝘢𝘵𝘪𝘰𝘯, 𝘛𝘦𝘴𝘵 𝘈𝘶𝘵𝘰𝘮𝘢𝘵𝘪𝘰𝘯, 𝘋𝘰𝘤𝘬𝘦𝘳, 𝘒𝘶𝘣𝘦𝘳𝘯𝘦𝘵𝘦𝘴, 𝘑𝘦𝘯𝘬𝘪𝘯𝘴, 𝘜𝘯𝘳𝘦𝘢𝘭 𝘌𝘯𝘨𝘪𝘯𝘦, 𝘎𝘪𝘵

𝐌𝐮𝐥𝐭𝐢-𝐓𝐞𝐧𝐚𝐧𝐭 𝐌𝐋 𝐋𝐚𝐛𝐞𝐥𝐢𝐧𝐠 & 𝐓𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐏𝐥𝐚𝐭𝐟𝐨𝐫𝐦

Built a scalable, multi-tenant ML platform handling labeling ingestion, data preparation, consensus computation, training orchestration, and real-time model updates for 20 custom categorization and entity recognition models across 4 game projects. Developed an event-driven shared label propagation service to efficiently update multiple ML models using related/similar labels, reducing human labeling time from 4 hours/week to 15–30 minutes while maintaining average model accuracy (~0.78). Reduced ML workflow runtime through parallel and multi-node orchestration.

𝙏𝙚𝙘𝙝 𝙃𝙞𝙜𝙝𝙡𝙞𝙜𝙝𝙩𝙨: 𝘈𝘞𝘚 𝘚𝘵𝘦𝘱 𝘍𝘶𝘯𝘤𝘵𝘪𝘰𝘯𝘴, 𝘓𝘢𝘮𝘣𝘥𝘢, 𝘉𝘢𝘵𝘤𝘩, 𝘍𝘢𝘳𝘨𝘢𝘵𝘦, 𝘌𝘊𝘙, 𝘚3, 𝘎𝘭𝘶𝘦, 𝘈𝘵𝘩𝘦𝘯𝘢, 𝘘𝘶𝘪𝘤𝘬𝘚𝘪𝘨𝘩𝘵, 𝘚𝘕𝘚, 𝘚𝘘𝘚, 𝘗𝘺𝘵𝘩𝘰𝘯, 𝘊#, 𝘌𝘷𝘦𝘯𝘵-𝘥𝘳𝘪𝘷𝘦𝘯 𝘢𝘳𝘤𝘩𝘪𝘵𝘦𝘤𝘵𝘶𝘳𝘦, 𝘋𝘪𝘴𝘵𝘳𝘪𝘣𝘶𝘵𝘦𝘥 𝘔𝘓 𝘸𝘰𝘳𝘬𝘧𝘭𝘰𝘸𝘴.