Summary
Overview
Work History
Education
Skills
Accomplishments
Languages
Hobbies and Interests
Docker Hub
Timeline
Hi, I’m

Dushyant Singh

Pune
Dushyant Singh

Summary

Results-driven AI & Data professional with 4+ years of experience delivering real world, business-impacting solutions. Skilled in full-stack AI workflows from data engineering and Data Science to LLM research and deployment. Passionate about emerging opensource models (DIA, Llama, Deepseek, Mistral, D-Fine), actively experimenting on Colab/HF Spaces while mentoring teams through deep learning and agentic AI bootcamps. Strong background in data management, communication and team leadership.

Overview

5
years of professional experience

Work History

FPL Technologies

AI Engineer
02.2025 - Current

Job overview

  • ChatBot intent classifier with language detection, translation & GPT fallback using Qdrant & Sentence Transformers, improved accuracy by 26% and cut costs by 44%.
  • Multilingual AI voice bot with Whisper ASR and Coqui TTS, integrated with telephony via WebRTC-VAD for low-latency calls; reduced manual effort by 30%.
  • Delivered a 6-week Deep-Learning & GenAI Bootcamp (PyTorch & TensorFlow) for 50+ engineers covering ANNs, CNNs, Transformers, LoRA fine-tuning, MCP Servers, AI Agents, Langchain, HF, cuML, WindSurf etc.
  • Weekly sentiment analysis of OneCard vs competitors (CRED, Uni) on Reddit & Twitter using keyword search. Used Hugging Face LLMs, Colab GPUs and X APIs.
  • Built LLM pipeline to analyze call remarks & suggest next-best actions for better collections, Fine-tuned Quantized Llama-3-8B to explain credit score changes in OneScore.

OneCard & OneScore

Data Management
03.2024 - 01.2025

Job overview

  • Led end-to-end data migration by organizing tables into structured L1/L2/L3 formats, leveraging PySpark for Big Data in AWS Glue Jobs to transform data into Delta tables.
  • Streamlined ETL pipelines by consolidating raw tables, eliminating redundant joins, and enhancing efficiency through collaboration with stakeholders to gather requirements.
  • Designed and implemented CI/CD pipelines using GitLab on EC2, S3 sync, and Lambda functions for cron job migrations on Starburst (Trino), enabling multi-source data access.
  • Guided an 80+ member Data team in query optimization, cost reduction, and workload migration to PySpark and Polars, ensuring seamless communication.
  • Led the AWS Feature Store project with Lambda-triggered pipelines for feature sharing across teams.
  • Implemented MLflow for model monitoring and governance, tracking stability metrics (KSI, PSI) on Superset without manual intervention and reporting discrepancies on Google Chat.
  • Enhanced PII data security by writing Glue jobs to identify PII tables and implementing Ranger-based masking for hash-based joins.
  • Automated workflows using Lambda and GitLab for Jupyter scripts, Glue jobs, SFTP uploads, and API calls, ensuring secure data processing.
  • Optimized Superset by migrating it to Trino, upgrading functionality with PDF exports, new charts, and GChat notifications.
  • Integrated OpenMetadata for data profiling, lineage tracking, test cases, and email alerts, while conducting training sessions to ensure team-wide adoption.
  • Built real-time data streaming pipelines using Amazon Kinesis, Firehose, and EventBridge.
  • Integrated Lambda functions within Firehose to preprocess streaming data, extracting relevant keys from callback tables for efficient storage, analysis.

FPL Technologies (OneCard)

Risk and Decision Scientist
11.2020 - 08.2023

Job overview

  • Developed XGBoost model for risk segmentation on 1 million Bucket 0 customers using Bureau, Transactions, App, Repayments data and achieved 81% Gini.
  • Treatment of High-risky customers improved resolution from 89% to 92.4% before due date.
  • Implemented Random Forest Classifier for predicting the best communication channel on 1.5M customers, achieving 83% Gini.
  • Minimized communication spamming while maintaining business growth. It helped in improving NPS by 7%.
  • Implemented risk Triggers, EMI, Autopay & Jackpot campaigns, optimized calling scripts and communication journey in Bucket0 which reduced entry rates from 3.18% to 2.54% on 3000+ Cr portfolio and flow by 20+Cr.
  • Developed XGBoost model (Gini 72%) to predict propensity to pay and stab logic for bucket-1 customers.
  • Allocated high risky accounts to best agencies which improved resolution by 13% on ~145Cr portfolio.
  • Automated various manual tasks on AWS such as Agency Allocation, Repayment & comms Dashboards, Pay-out Emails, FLDG, Model Vars etc. resulted in increased efficiency and reduced manual intervention.
  • Developed Decision Tree model utilizing previous disposition and calling variables to predict repayment capability for Bucket 2-3.
  • Applied model in agency allocation and implemented stab-rollback logic, leading to reduced flow into NPA, 9% lower losses, and 0.23% decrease in POF.
  • Worked with sourcing team and took action to minimize credit losses through analyzing sourcing cohorts, Bank Migration, New cities Launch, Bureau delinquency and identification of frauds by high-risk patterns.
  • Finalized optimal billing dates and ran campaigns for 5 new banks( resulted in same entry rates 2.6%), new LPC matrix (Increased LPC revenue by ~30%), Payout grid & Incentives, Risk based comms journeys (improved DIY by 3%).
  • Collaborated with multiple teams to develop an internal collection system OneAccord, Integrated it with Dialer, Nudge, Billing and contact DB enabling data security and improved productivity of internal team.

Education

Indian Institute of Technology BHU

B-Tech from Civil Engineering
04.2020

Skills

  • GenAI
  • Agentic AI
  • Transformers
  • RAG
  • HuggingFace
  • CrewAI
  • Google ADK
  • LangChain
  • LlamaIndex
  • Ollama
  • Encoder-Decoder
  • Attention Mech
  • MCP Server
  • Vector DBs
  • Finetuning
  • Quantization
  • Streamlit
  • Python
  • SQL
  • PySpark
  • XGBoost
  • Random Forest
  • RNN
  • CNN
  • ANN
  • PyTorch
  • TensorFlow
  • Risk Modeling
  • Statistical analysis
  • DoWhy
  • SPSS
  • Excel
  • AWS Glue
  • Kinesis
  • EC2
  • Lambda
  • ETL
  • Docker
  • Gitlab CI/CD
  • Superset
  • Data Pipelines
  • GChat Alerts
  • Cost optimization
  • MLflow
  • Feature Store
  • Bedrock
  • Ranger
  • OpenMetadata
  • Lineage
  • Problem-Solving
  • Leadership
  • Collaboration
  • Data Storytelling
  • Product Development
  • OneAccord
  • Starburst

Accomplishments

  • One Accord Award, FPL Technologies, Successfully completed the OneAccord project within two months, closely coordinating with team members and detailed planning.
  • Honorable Mention, Gymkhana Award, Contribution in the growth of Film and Media Council at IIT BHU Varanasi.

Languages

English

Hobbies and Interests

  • Photography
  • Hiking
  • Swimming
  • Music
  • Drone Videos

Docker Hub

hub.docker.com/repositories/dushyant1334

Timeline

AI Engineer

FPL Technologies
02.2025 - Current

Data Management

OneCard & OneScore
03.2024 - 01.2025

Risk and Decision Scientist

FPL Technologies (OneCard)
11.2020 - 08.2023

Indian Institute of Technology BHU

B-Tech from Civil Engineering

Similar Profiles

CREATE PROFILE
Dushyant Singh