GenAI Engineer with 9+ Years Experience

Sai Keerthi

Phone : +1-904-***-****

Email: *******************@*****.***

Gen AI Engineer

PROFESSIONAL SUMMARY:

Generative AI Engineer with 9+ years of experience delivering enterprise-scale AI/ML and LLM-driven systems across healthcare and financial services domains.

Specialized in architecting production-grade GenAI pipelines in regulated, high-volume environments. Proven expertise in full AI lifecycle ownership from architecture to deployment and monitoring.

Deep expertise in Retrieval-Augmented Generation (RAG), embedding architectures, and semantic knowledge systems, designing high-performance vector search platforms using FAISS, Pinecone, Weaviate, and ChromaDB.

Built scalable AI systems enabling contextual reasoning across structured and unstructured enterprise data.

Extensive experience working with GPT, LLaMA, Claude, and AWS Bedrock, building enterprise copilots, AI agents, and intelligent automation workflows.

Leveraged LangChain and LlamaIndex to orchestrate multi-step reasoning pipelines and fine-tuned LLMs for performance, latency optimization, and cost efficiency.

Architected and deployed multi-cloud GenAI solutions across AWS, Azure, and GCP, utilizing SageMaker, Bedrock, Vertex AI, Azure ML, and Dataflow.

Designed secure API-driven integrations with OAuth 2.0, RBAC, and enterprise authentication protocols, ensuring compliance-aligned AI services.

Strong expertise in prompt engineering, chain-of-thought reasoning, structured output generation, and LLM evaluation frameworks.

Implemented guardrails, bias mitigation, hallucination reduction, and inference optimization using GPU acceleration, distributed training, and quantization techniques.

Led enterprise-scale MLOps and LLMOps initiatives, automating model lifecycle management using MLflow, Kubeflow, Jenkins, GitLab CI/CD, Docker, and Kubernetes.

Established monitoring, drift detection, and observability pipelines with Prometheus and Grafana for reliable AI deployments.

Delivered high-impact GenAI-powered solutions including fraud assistants, healthcare document processors, and financial copilots, while building advanced ML systems (XGBoost, CNNs, RNNs, GANs, VAEs, Prophet, ARIMA, LSTM). Known for translating complex AI research into scalable, business-ready enterprise platforms.

TECHNICAL SKILLS:

Generative AI & NLP

LLMs ( Llama3), RAG 2.0 Architectures, Retrieval Grounding & Guardrails, Lang Chain, Lang Graph, Llama Index, DSPy, Lang Smith,Stateful Agents, Multi-Agent Systems, Tool Calling, MCP, Prompt Engineering & Context Optimization, Fine-Tuning (LoRA, PEFT, QLoRA), Embedding Optimization,Vector Databases (Pinecone, Weaviate, FAISS).

Machine Learning & Deep Learning

PyTorch, TensorFlow, Keras, scikit-learn, XGBoost, LightGBM, Supervised & Unsupervised Learning, Anomaly Detection, Feature Engineering, Model Evaluation, Cross-Validation, Ensemble Methods, Autoencoders, Graph Neural Networks (GNNs), Statistical Modeling, Optimization Techniques.

Cloud & Infrastructure

AWS (Bedrock, SageMaker, Lambda, EC2, S3, Redshift, Glue, Step Functions, API Gateway, CloudFormation, CloudWatch), Azure (Azure OpenAI, ML Studio, Synapse, DevOps CI/CD), GCP (Vertex AI, BigQuery, AutoML, Dataflow), Terraform, Docker, Kubernetes (EKS), Autoscaling, VPC Networking

MLOps & Production Engineering

MLflow, Kubeflow, Airflow, Cloud Composer, CI/CD (GitHub Actions, GitLab CI, Jenkins), Model Monitoring (Prometheus, Grafana, Splunk, Kibana), Drift & Bias Detection, Automated Retraining Pipelines, Experiment Tracking, Model Registry, Canary Deployments

Software Engineering & APIs

Fast API, REST APIs, Flask, Git, Python, SQL, JavaScript, Hugging Face, NumPy, Pandas, SciPy

Databases & Storage

PostgreSQL, MongoDB, DynamoDB, Redshift, Snowflake, Big Query, HBase, Vector Databases (Pinecone, FAISS, Weaviate, Milvus), Data Warehousing (Star Schema, ETL/ELT pipelines)

Visualization & Statistical Analysis

Tableau, Power BI, Matplotlib, Seaborn, Plotly, A/B Testing, PCA, Clustering (KMeans, DBSCAN), Regression Analysis, ANOVA, Dimensionality Reduction, Cross-Validation, Hypothesis Testing, Bayesian Inference.

Tools & Methodologies

Anaconda, Flask, Django, OAuth 2.0, SDLC (Agile/Scrum), TDD/BDD

PROFESSIONAL EXPERIENCE:

UnitedHealth Group Minnetonka, Minnesota Jul 2024 - Present

Gen AI Engineer

Responsibilities:

As a GenAI Engineer, designed and deployed scalable AWS-native GenAI + ML fraud detection systems across regulated, enterprise-scale healthcare claims environments.

The initiative targeted financial leakage from upcoding, unbundling, phantom billing, and utilization anomalies by embedding ML + LLM intelligence into SIU investigation workflows.

Reduced investigators review time by 50%+ and improved fraud detection accuracy by 30% through structured ML scoring integrated with Bedrock-hosted LLM reasoning pipelines.

Architected multi-stage ML + LLM workflows integrating CPT/ICD codes, claim narratives, provider risk signals, and historical SIU outcomes into unified fraud intelligence pipelines.

Built high-throughput ingestion pipelines using AWS Glue, MSK (Kafka), EMR (PySpark), S3, and Redshift to process millions of healthcare claims for anomaly detection and retrieval-based reasoning.

Implemented RAG 2.0 architectures using LangChain, LangGraph, and DSPy with AWS Bedrock to enable contextual policy reasoning grounded in payer rules and medical guidelines.

Developed Bedrock-based reasoning modules leveraging function-calling and controlled decoding to generate deterministic JSON fraud signals for downstream SIU systems.

Led architectural governance and AI design reviews, defining standardized GenAI patterns for RAG orchestration, ML + LLM workflow isolation, and enterprise fraud analytics scalability.

Containerized AI microservices using Docker and deployed them on Amazon EKS with autoscaling, canary releases, and high-availability configurations for real-time fraud scoring.

Exposed fraud-risk scoring, claim summarization, and provider-risk assessment services through FastAPI endpoints integrated into enterprise investigation platforms.

Automated retraining workflows using SageMaker Pipelines, MLflow, and Airflow, incorporating adjudicated claims and SIU feedback loops for continuous model refinement.

Optimized LLM inference cost and latency through request batching, token optimization, and Bedrock configuration tuning, reducing compute expenditure while maintaining reasoning reliability.

Established production observability using CloudWatch, Prometheus, OpenSearch, and MLflow to monitor drift, anomaly spikes, latency deviations, and scoring degradation.

Built automated drift detection and retraining triggers to ensure fraud models continuously adapted to evolving provider behaviors and emerging fraud schemes.

Integrated SHAP and LIME explainability frameworks to generate transparent fraud-risk justifications, improving investigator trust and audit readiness.

Enforced HIPAA-aligned AI deployment practices using IAM, KMS encryption, VPC isolation, RBAC controls, and CloudTrail audit logging for secure PHI handling.

Evaluated LLM reasoning quality using LangSmith and TruLens, implementing retrieval grounding and hallucination-reduction guardrails to ensure reliable outputs.

Introduced multimodal and multi-agent LLM architectures to analyze medical documentation, validate billing logic, and escalate high-risk claims through structured AI reviewer chains.

Environment: AWS Bedrock, SageMaker, EKS, S3, Redshift, Glue, EMR, Kafka (MSK), LangChain, LangGraph, DSPy, PyTorch, FastAPI, Docker, MLflow, Airflow, Terraform, SHAP, LIME.

Merck Kenilworth, New Jersey Apr 2022 - Jun 2024

AI/ML Engineer

Responsibilities:

As an AI/ML Engineer at Merck, designed and deployed Azure-native ML pipelines supporting molecule screening, QSAR modeling, ADMET prediction, and hit-to-lead prioritization across regulated pharmaceutical R&D environments.

The primary objective was to accelerate drug discovery cycles by improving toxicity prediction accuracy, compound prioritization efficiency, and model interpretability across high-throughput molecular datasets.

Improved compound prioritization lift and reduced molecular screening turnaround time by optimizing distributed training pipelines and GPU-based workloads on Azure ML compute clusters.

Architected scalable machine learning workflows using Azure Machine Learning, Azure Databricks, and Azure Synapse Analytics to process high-volume assay datasets and molecular descriptor libraries.

Built end-to-end QSAR modeling pipelines leveraging RDKit, DeepChem, PyTorch, and XGBoost, integrated with Azure ML experiment tracking and automated model versioning.

Implemented Azure Data Factory and Databricks-based ingestion frameworks to manage structured and semi-structured molecular datasets stored in ADLS Gen2.

Optimized distributed GPU training strategies on Azure ML and AKS, reducing computational bottlenecks and improving large-scale molecular modeling throughput.

Developed literature intelligence pipelines using Hugging Face Transformers for scientific document analysis and integrated Azure OpenAI APIs for contextual knowledge extraction, later extending the architecture with LangChain-based orchestration workflows to enable advanced retrieval and reasoning across research publications.

Containerized predictive services using Docker and deployed scalable inference APIs on Azure Kubernetes Service (AKS), enabling real-time molecular property predictions for R&D applications.

Automated retraining and experimentation workflows using Azure ML Pipelines, MLflow, and Azure DevOps CI/CD, ensuring reproducibility and regulatory audit readiness.

Applied SHAP and LIME explainability integrated with Azure ML interpretability tools to identify molecular substructures influencing ADMET predictions.

Collaborated with medicinal chemists, computational biologists, and research operations teams to embed predictive outputs into interactive compound design dashboards powered by Power BI.

Implemented structured evaluation frameworks including cross-validation, hyperparameter tuning, and benchmarking across classical ML, Graph Neural Networks, and transformer-based chemical encoders.

Conducted innovation initiatives leveraging Azure OpenAI and transformer-based chemical encoders to explore hybrid modeling approaches combining structural descriptors with deep learning embeddings.

Environment: Azure Machine Learning, Azure OpenAI, Azure Databricks, Azure Synapse Analytics, Azure Data Factory, ADLS Gen2, Azure Kubernetes Service (AKS), Azure DevOps, MLflow, PyTorch, DeepChem, RDKit, XGBoost, Graph Neural Networks, Hugging Face Transformers, LangChain, Power BI, GPU-Accelerated Distributed Training.

Truist Bank Charlotte, NC Aug 2020 - Mar 2022

Machine Learning Engineer

Responsibilities:

As a Machine Learning Engineer at Truist, led the design and implementation of real-time fraud detection systems across digital banking, ACH, card, and Zelle payment channels, owning the end-to-end ML lifecycle from requirements through production deployment within enterprise financial systems.

The primary business objective was to reduce fraud losses while minimizing customer-impacting false positives, defining KPIs around precision/recall, sub-second latency, and investigator efficiency while aligning ML solutions with regulatory and operational risk requirements.

Delivered measurable impact by improving fraud detection accuracy, reducing false positives, and enabling sub-second transaction scoring across millions of daily transactions, significantly accelerating investigator workflows and strengthening audit readiness.

Architected a scalable AWS-based fraud detection platform using SageMaker, S3, Glue, Redshift, and Lambda designing distributed ingestion pipelines capable of near real-time scoring while ensuring high availability and security.

Built advanced feature engineering frameworks generating device-level risk signals, transaction velocity metrics, geo-deviation indicators, and behavioral fingerprints, structuring reusable feature stores to support consistent training and inference.

Developed supervised and unsupervised anomaly detection models including XGBoost, Random Forest, Isolation Forest, Logistic Regression, and Autoencoders, optimizing hyperparameters to improve detection precision while reducing customer friction.

Built transformer-based NLP pipelines using BERT and FinBERT to analyze transaction narratives and implemented document retrieval and contextual search frameworks using Elasticsearch, enabling investigators to quickly identify related transaction evidence and generate contextual fraud summaries.

Optimized distributed model training using PySpark and MLlib across multi-million-record datasets, improving data partitioning and Spark orchestration strategies and reducing training time by 35%.

Containerized fraud detection services using Docker and deployed scalable inference workloads on Kubernetes (EKS), exposing FastAPI microservices to support low-latency, real-time fraud scoring integrations.

Implemented CI/CD automation using Bamboo, Git, and MLflow to streamline model versioning, promotion, and rollback strategies, improving release reliability in high-risk production environments.

Established production monitoring and observability pipelines using MLflow dashboards and AWS logging frameworks to track precision, recall, drift patterns, anomaly spikes, and latency deviations in live systems.

Built automated drift detection and retraining triggers to ensure fraud models adapted to evolving attack patterns, implementing validation checks and alerting mechanisms to maintain scoring integrity.

Applied SHAP and LIME explainability techniques to generate transparent, audit-ready fraud-risk justifications aligned with financial regulatory standards, improving trust in model-driven decisions.

Collaborated cross-functionally with risk analytics, cybersecurity, compliance, and DevOps teams to align fraud detection strategies with operational workflows and regulatory mandates, strengthening enterprise adoption and governance.

Environment: AWS (SageMaker, Lambda, Glue, Redshift, S3), Kubernetes (EKS), Docker, PySpark, MLlib, Snowflake, FastAPI, MLflow, Bamboo CI/CD. XGBoost, Isolation Forest, Autoencoders, BERT, FinBERT, GPT, Retrieval-Augmented Generation (RAG), SHAP, LIME, Model Drift Detection, Real-Time Inference Architecture.

Home Depot - Atlanta, Georgia Jun 2019 - Jul 2020

Data Scientist

Responsibilities:

As a Data Scientist at Home Depot, led end-to-end demand forecasting and personalization initiatives across supply chain and digital commerce divisions, designing cloud-native ML pipelines to optimize enterprise retail operations.

The primary business objective was to improve forecast accuracy, reduce inventory volatility, and enhance digital customer engagement by aligning ML solutions with turnover ratios and margin optimization KPIs.

Delivered measurable impact by improving sales forecast accuracy by 25%, reducing overstock/stockout scenarios, and enabling data-driven merchandising decisions across multi-region retail networks.

Architected scalable ML systems for time-series forecasting, recommendation engines, and NLP-driven sentiment analytics, establishing structured ingestion-to-deployment workflows across enterprise Azure infrastructure.

Built advanced forecasting models (Random Forest, XGBoost, ARIMA, Prophet, LSTM) and collaborative filtering–based recommendation systems to drive SKU-level demand prediction and personalized product engagement.

Designed NLP pipelines using spaCy and BERT-based sentiment models to analyze customer feedback and integrate sentiment signals into demand forecasting and merchandising analytics workflows.

Developed modular data pipelines using Python, PySpark, SQL, and Airflow, automating feature engineering, retraining workflows, and model versioning to ensure reproducibility and continuous data refresh cycles.

Optimized distributed processing across Azure Synapse and Databricks using Spark and Hadoop, reducing big-data processing runtimes and improving scalability for multi-terabyte retail datasets.

Implemented rigorous validation frameworks including cross-validation, hyperparameter tuning, statistical hypothesis testing, A/B testing, and applied SHAP/LIME explainability to enhance model transparency and stakeholder trust.

Deployed containerized ML inference services using Docker and Kubernetes, integrated via REST APIs into enterprise inventory systems, and established monitoring and drift detection pipelines using MLflow and Azure monitoring tools.

Environment: Azure Machine Learning, Azure Synapse Analytics, Azure SQL Database, Azure Databricks, Python, XGBoost, LSTM (Deep Learning), ARIMA, Prophet, Random Forest, Neural Networks, BERT, spaCy, Hugging Face Transformers, Apache Spark, Hadoop, PySpark, MLflow, DVC, Apache Airflow, CI/CD Pipelines, Docker, Kubernetes, REST APIs, SHAP, LIME, Power BI.

Edvensoft Solutions India Pvt. Ltd India Jun 2016 - Feb 2019

Data Scientist

Responsibilities:

As a Data Scientist at Edvensoft, led end-to-end AI initiatives across retail and finance domains, designing scalable machine learning architectures that translated complex business problems into production-ready analytics and intelligent automation systems.

The primary objective was to improve anomaly detection, operational forecasting, and decision intelligence by building structured data foundations capable of supporting scalable AI-driven insights across enterprise environments.

Delivered measurable impact by improving anomaly detection efficiency by 25%, accelerating operational response times, and establishing a scalable analytics foundation that later evolved into advanced AI and GenAI-enabled systems.

Architected large-scale data ingestion and transformation pipelines using Hadoop (HDFS, MapReduce), Hive, and HBase, creating distributed data processing frameworks aligned with modern large-scale AI training and inference architectures.

Built modular ML frameworks for anomaly detection, segmentation, and forecasting using Random Forest, SVM, k-means, and Logistic Regression, establishing reusable experimentation pipelines aligned with structured model lifecycle practices.

Developed automated ETL and feature engineering workflows using Python, SQL, Hive, and Airflow, enabling reproducible data pipelines and structured datasets suitable for downstream AI-assisted decision systems and retrieval-based workflows.

Implemented hybrid anomaly detection systems combining rule-based logic with probabilistic ML outputs, an approach foundational to AI guardrails, model explainability, and reliability frameworks used in modern GenAI systems.

Established strong validation, monitoring, and governance practices including cross-validation, hypothesis testing, drift detection, and BI integration (Tableau, Power BI), ensuring transparency, performance benchmarking, and enterprise-grade AI governance.

Environment: Hadoop (HDFS, MapReduce), Hive, HBase, Apache Airflow, Python, SQL, R, Scikit-learn, Random Forest, SVM, k-means, TensorFlow 1.x, Feature Engineering, Statistical Validation, Tableau, Power BI, Linux.

Contact this candidate