Dushyant Singh

Multilingual ChatBot intent classifier with Qdrant, improved accuracy by 8% reducing 175+ daily customer support calls. Delivered 12 DL + GenAI sessions for 50+ FPL engineers covering ANNs, AI Agents, LoRA, MCP, Transformers, CNNs, Langchain, HF, cuML & TensorFlow. Deployed AI voicebot (Whisper, Coqui TTS) to handle ~30% of the outbound calls. Created a weekly Reddit/X sentiment tracker for OneCard vs competitors, finding key insights & roadmap updates that boosted positive mentions by 16%. Made a LLM pipeline to analyze call remarks & suggest next best actions which raised resolution rates by 2.2% on 400Cr+ portfolio.

Migrated 100+ TB of 40M+ customers data to structured Delta Lakes, optimized ETLs and eliminating redundant joins, reduced compute costs by ~15%. Implemented CI/CD on EC2 with GitLab and S3 for cron jobs. Led 80+ team in optimizing queries and migrating to PySpark/Polars. Designed real-time pipelines with AWS Kinesis, Firehose and Lambda to process crucial data with sub-second latency, lifted risk analysis efficiency by 25%. Secured PII with Glue + Ranger masking and automated data workflows using SFTP uploads & API calls to ensure RBI compliance. Upgraded Superset with Trino (enhanced dashboard load times by 40% & reduced cost), charts and OpenMetadata for data profiling, alerts. Led AWS Feature Store with Lambda pipelines. Automated ML monitoring via MLflow, Superset and GChat alerts (caught anomalies 2x faster).

Risk model(XGBoost, 81% Gini), boosted repayments by 3.4% on 3000Cr+ portfolio. Cut 20Cr+ flow by optimizing campaigns, triggers and EMI journeys on Bucket-0. Launched internal collection system OneAccord, boosted calling team productivity by ~30%. Developed risk model for bucket-1, improving resolution by 13% on 145Cr. Automated key tasks like allocation, dashboards and payouts to cut manual work. Worked with sourcing & fraud team to cut losses before new cities/bank launch.