Jay
Krishna
AI/ML Engineer —
8+ years designing and deploying production-grade AI/ML systems, Generative AI pipelines, and cloud data platforms for Wells Fargo, Walmart, and United Health Group.
About Me
I am an AI/ML Engineer with 8+ years of experience designing and deploying scalable, production-grade AI/ML and data solutions across cloud and enterprise environments. I have full ownership of the ML lifecycle — from data ingestion and feature engineering through model training, deployment, monitoring, and retraining.
My expertise spans Generative AI (RAG, LLM fine-tuning, LangChain, LangGraph, Azure OpenAI), deep learning (CNNs, LSTMs, Transformers, BERT, GPT-4), and classical ML (XGBoost, LightGBM, Random Forest, Isolation Forest). I build and deploy AI microservices with FastAPI and Docker on Kubernetes (AKS/EKS).
I have delivered end-to-end AI/ML workloads on AWS SageMaker and Azure ML, integrated real-time inference with Apache Kafka and Spark Streaming, and implemented full MLOps pipelines using MLflow, DVC, GitHub Actions, and Terraform.
Strong background in Python, PySpark, Scala, and the Hadoop ecosystem, with hands-on experience in vector databases (FAISS, Pinecone) for semantic search and recommendation systems.
Technical Skills
A production-grade AI/ML stack built over 8+ years across finance, retail, and healthcare.
Work Experience
8+ years across financial services, retail, healthcare, and IT consulting.
- Designed end-to-end ML lifecycle covering data ingestion, feature engineering, model training, real-time inference, and monitoring using PySpark pipelines in Azure Databricks, orchestrated through Azure Data Factory.
- Built fraud classification models using XGBoost and Random Forest on transaction amount, merchant category, geolocation, and device fingerprint features; implemented real-time anomaly detection using Isolation Forest and Autoencoder on streaming data.
- Developed time-series models using LSTM and ARIMA to detect seasonal fraud trends and predict high-risk transaction windows.
- Built a RAG-based fraud investigation assistant using FAISS and HuggingFace SentenceTransformers to help analysts retrieve historical cases and investigation playbooks; orchestrated multi-step reasoning with LangGraph.
- Containerized and deployed ML scoring services via FastAPI, Docker, and Azure Kubernetes Service (AKS) with autoscaling for sub-second latency fraud scoring.
- Implemented model explainability using SHAP and LIME for regulatory audit compliance under BSA/AML guidelines.
- Managed experiment tracking and model versioning using Azure ML integrated with MLflow; developed Prometheus and Grafana dashboards to monitor fraud scoring latency and data drift.
- Designed end-to-end ML pipelines for data ingestion, feature engineering, model training, batch scoring, and monitoring using PySpark on Azure Databricks, orchestrated through Apache Airflow.
- Built demand forecasting models using XGBoost, LightGBM, and ARIMA to predict product demand patterns, reducing inventory costs and improving stock availability across regions.
- Implemented customer segmentation using K-Means clustering on large-scale POS transaction data to support targeted marketing strategies.
- Engineered features from retail transaction data including rolling aggregates and seasonal indicators using Spark window functions; applied SHAP for interpretable feature importance insights.
- Deployed containerized ML scoring services with Docker; orchestrated CI/CD pipelines using Jenkins and Git for reproducible deployments.
- Implemented Kafka and Spark Streaming for live POS data analytics; developed Power BI dashboards for KPI tracking and executive reporting.
- Migrated pipeline orchestration from Oozie to Airflow; recreated application logic in Azure Data Lake and Azure Data Factory.
- Used Azure Data Factory extensively for ingesting healthcare data from disparate source systems, automating jobs using Event, Scheduled, and Tumbling Window triggers.
- Created numerous ADF v2 pipelines with Copy, Filter, ForEach, and Databricks notebook activities for end-to-end data movement and transformation.
- Provisioned Databricks clusters for batch and streaming workloads; developed PySpark jobs for complex table-to-table operations and data transformations.
- Ingested data in mini-batches and performed RDD transformations using Spark Streaming for streaming analytics in Databricks.
- Integrated Azure Active Directory authentication to Cosmos DB requests; created Build and Release definitions for CI/CD using Azure DevOps.
- Authored high-level technical design and application design documents; worked with complex SQL, Stored Procedures, and Triggers across large databases.
- Administered and maintained Cloudera Hadoop clusters on Linux environments; managed cluster coordination via Zookeeper.
- Wrote multiple MapReduce programs for data extraction, transformation, and aggregation from 20+ sources across XML, JSON, and CSV formats.
- Created Hive external tables, loaded data, and performed HQL queries for ad-hoc analytics; created Oozie workflows orchestrating Sqoop imports and Hive scripts.
- Worked in AWS (EC2, S3) for deployment of custom Hadoop applications; transferred data from S3 to Redshift using Informatica.
- Utilized Python Matplotlib and Scikit-Learn for prototype visualizations and early ML experiments; generated business reports using SSRS.
- Performed data validation using MapReduce by building custom models to filter invalid records and cleanse datasets.
Key Projects
Highlights from enterprise AI/ML and data engineering work across multiple industries.
Built an end-to-end fraud detection system analyzing millions of transactions in real time using XGBoost, Isolation Forest, and LSTM models on Azure Databricks with Kafka streaming. Deployed as FastAPI microservices on AKS with sub-second latency.
Built a Generative AI investigation assistant using RAG (FAISS + HuggingFace SentenceTransformers) and LangGraph multi-step reasoning to help fraud analysts retrieve historical cases and regulatory guidelines. Integrated GPT-4 via Azure OpenAI Service.
Built demand forecasting models using XGBoost, LightGBM, and ARIMA to predict product demand across regions. Implemented K-Means customer segmentation on POS data to support targeted marketing. Managed full MLflow experiment tracking on Databricks.
Designed end-to-end ML pipelines on Azure Databricks orchestrated via Apache Airflow. Deployed containerized scoring services with Docker and Jenkins CI/CD. Implemented Kafka + Spark Streaming for live POS analytics and Power BI dashboards for leadership.
Built large-scale ETL workflows on Azure to process healthcare data from multiple source systems. Used ADF v2 with Databricks notebooks, Spark Streaming for mini-batch ingestion, and Azure DevOps CI/CD. Integrated Cosmos DB with Azure AD authentication.
Administered Cloudera Hadoop clusters on Linux and wrote MapReduce jobs to ETL data from 20+ heterogeneous sources. Used Scikit-Learn for early ML prototypes; transferred data between HDFS, AWS S3, and Redshift via Sqoop and Informatica.
Education & Certifications
Academic foundation backed by years of hands-on AI/ML enterprise practice.
Get In Touch
Available for new AI/ML engineering roles, consulting, and collaborations.
Let's Work Together
Available for full-time roles, contract work, or consulting engagements in AI/ML engineering and cloud architecture.
Send a Message
Fill out the form below and I'll get back to you promptly.