At CoinDCX, I designed an Announcement Tracking System to triage daily TP announcements for TradeOps. Across four iterations, it reached 98% recall and 70% precision, cutting manual review effort by about 75%.
I also built a compensation pipeline mapping CS and TradeOps workflows into Databricks, validating it against more than 3,000 historic requests. I standardized the pre-automation intake process, which cut blockers by 35% on the next initiative.
Previously, at Stylumia, I benchmarked forecasting models and built an LLM evaluation framework that raised the trend recommendation hit rate from 0/10 to 8/10. At GeeksforGeeks, I designed a recommendation engine that improved retention by 15% and CTR by 25%.
