GenAI Engineer with 10+ Years Enterprise AI Delivery

VENUGOPAL REDDY

*****.*************@*****.*** +1-872-***-**** linkedin.com/in/94venugopal

PROFESSIONAL SUMMARY

GenAI Engineer with over 10 years of experience delivering enterprise-scale AI, ML, and LLM solutions across insurance, healthcare, banking, retail, and digital product ecosystems.

Expert in modern LLM development, including GPT-4o/GPT-4, LLaMA, Mistral, Claude, and domain-tuned transformer models, with hands-on experience in LoRA/QLoRA, PEFT, Hugging Face Transformers, prompt engineering, and instruction tuning for specialized reasoning tasks.

Strong in Retrieval-Augmented Generation (RAG) using Azure Cognitive Search, FAISS, ChromaDB, and Amazon OpenSearch, with hybrid, semantic, and sliding-window chunking strategies to build accurate, high-relevancy retrieval pipelines.

Designed multi-agent GenAI systems using LangChain, LangGraph, and MCP for claim automation, FNOL interpretation, clinical review workflows, and fraud investigation, enabling structured tool use and domain-aware decision paths.

Built scalable ML and GenAI microservices using FastAPI, Docker, Kubernetes (AKS/EKS/GKE), Pydantic, and CI/CD (GitHub Actions, Terraform, Jenkins) for secure, cloud-native deployment across Azure, AWS, and GCP.

Developed advanced machine learning and NLP models for fraud detection, risk scoring, forecasting, and document intelligence using PyTorch, TensorFlow, XGBoost, LightGBM, CatBoost, BERT/BioClinicalBERT, CNNs, RNNs, and NLP libraries including spaCy and NLTK.

Extensive MLOps experience, implementing MLflow, SageMaker Pipelines, Vertex AI Pipelines, model monitoring, drift detection, automated retraining, and governance workflows in highly regulated industries.

Implemented multimodal and document-intelligence pipelines integrating OCR, Form Recognizer, CNNs, and LLM-based extraction for ACORD forms, clinical notes, police reports, invoices, and financial narratives.

Strong cloud engineering foundation across Azure (AKS, Functions, Cosmos DB), AWS (SageMaker, Lambda, Step Functions, Fargate), and GCP (Vertex AI, BigQuery, Dataflow), building reliable, production-grade AI platforms.

Collaborative problem solver, partnering with claims teams, underwriters, clinicians, risk analysts, and product owners to convert complex business workflows into scalable, measurable AI and GenAI solutions.

TECHNICAL SKILLS:

Languages

Python, SQL, JavaScript, Bash, R

LLMs and GenAI

Azure OpenAI (GPT-4o/4/3.5), Amazon Bedrock, LLaMA, Mistral, Claude, Hugging Face Transformers, PEFT, LoRA/QLoRA, Instruction Tuning, Prompt Engineering, MCP (Model Context Protocol)

RAG and Vector Search

Azure Cognitive Search, FAISS, ChromaDB, Amazon OpenSearch, Weaviate, Hybrid Search, Context Compression, Embedding Pipelines, Retrieval Evaluation

Frameworks

LangChain, LangGraph, LlamaIndex, LCEL, FastAPI, Flask, Pydantic, Pytest, Neo4j (GDS), spaCy, NLTK, SentenceTransformers, Postman, Insomnia

Modeling

PyTorch, TensorFlow, Keras, Scikit-learn, XGBoost, LightGBM, CatBoost, BERT, CNNs, RNN/LSTM/GRU, Autoencoders, summarization, sentiment

MLOps and APIs

FastAPI, Docker, Kubernetes AKS, MLflow, GitHub Actions, Terraform, model versioning, drift detection, AB testing.

Cloud Platforms

Azure, Azure Databricks, AWS SageMaker, Lambda, Step Functions, ECS Fargate, Glue, S3, OpenSearch, EC2, CloudWatch, Vertex AI, BigQuery, Pub/Sub, Dataflow, GKE, Cloud Composer

Databases & Storage

Azure Synapse, Amazon Redshift, PostgreSQL, SQL Server, MongoDB, Cosmos DB, Chroma DB, Neo4j, BigQuery, Amazon S3

Monitoring & Responsible AI

App Insights, Azure Monitor, CloudWatch, Prometheus, Grafana, guardrails, OpenAI moderation, SHAP, LIME.

Visualization & BI

Tableau, Power BI, Streamlit, Splunk

Tools & DevOps

Git, GitHub, GitLab, JIRA, Confluence, Github Copilot

PROFESSIONAL EXPERIENCE

Liberty Mutual GenAI Engineer May 2024 - Present Boston, MA

Roles & Responsibilities:

Designed multi-agent GenAI workflows using Azure OpenAI, LangChain, LangGraph, and MCP, enabling Planner, Retriever, Reasoning, and Compliance agents and improving task-resolution reliability by 14% across coverage-analysis scenarios.

Built a production RAG pipeline with Azure Cognitive Search (BM25 + vector), FAISS, and ChromaDB to retrieve policy clauses, endorsements, and prior-claim data, improving retrieval quality on internal evaluation sets.

Developed loss-summary generation modules using GPT-4o and LangGraph to extract structured facts from accident statements, repair estimates, adjuster notes, and subrogation documents, enabling faster claim-review cycles.

Fine-tuned LLaMA and Mistral models with LoRA/QLoRA on de-identified insurance datasets, increasing structured-response accuracy by 12% while lowering GPU utilization during training cycles.

Engineered FNOL interpretation pipelines with multi-step agentic reasoning, conditional tool-calling, and retrieval orchestration to standardize extraction of insured parties, incident attributes, liability hints, and deductible logic.

Configured Cosmos DB with vector indexing to store embeddings, claim metadata, and agent-conversation state, reducing lookup and retrieval latency in multi-turn workflows.

Built document-intelligence pipelines using Form Recognizer, OCR, and LLM-based extraction to normalize ACORD forms, invoices, police reports, medical summaries, and handwritten adjuster notes.

Developed Azure Functions micro-pipelines for claim-type classification, coverage-trigger detection, document validation, and workflow routing into policy-administration and SIU systems.

Containerized GenAI microservices with Docker and deployed them on AKS using optimized CPU/GPU autoscaling profiles to maintain stable inference performance during peak claim-processing periods.

Delivered internal GenAI APIs (RAG, summarization, policy lookup, fraud-indicator scoring) using FastAPI + Pydantic for strict schema validation and validated all endpoints with Postman and Insomnia.

Provisioned cloud infrastructure with Terraform, including Cognitive Search indexes, Cosmos DB containers, AKS clusters, Key Vault integration, private networking, and managed identities for secure, reproducible deployments.

Evaluated agent reliability using LangSmith and custom harnesses, identifying and resolving 15% of tool-call inconsistencies and retrieval-routing failures.

Optimized prompts and retrieval parameters, tuning window sizes, chunking logic, reasoning depth, temperature, and context windows to reduce token usage by 11% while preserving output quality.

Built Streamlit and Power BI dashboards to visualize retrieval relevancy, agent-action flows, latency trends, and service-health metrics, supporting rapid debugging and operational monitoring.

Environment: Azure OpenAI (GPT-4o), LLaMA, Mistral, LangChain, LangGraph, MCP, FAISS, ChromaDB, Azure Cognitive Search, Cosmos DB (vector), Form Recognizer, Azure Databricks, Azure Functions, AKS, Docker, Terraform, FastAPI, Pydantic, LangSmith, Postman, Insomnia, Streamlit, Power BI, GitHub Copilot.

HCSC Senior Data Scientist / Gen AI Engineer Jan 2022 - May 2024 Chicago, IL

Roles & Responsibilities:

Developed clinical risk prediction models using XGBoost, LightGBM, and deep neural networks for readmission and disease-progression forecasting, improving targeting accuracy by approximately 25% on internal evaluation datasets.

Implemented GenAI-driven clinical summarization pipelines using Amazon Bedrock, LangChain, and retrieval-enhanced prompting, enabling clinicians to review multi-encounter patient histories more efficiently during pilot testing.

Built RAG workflows with FAISS, Amazon OpenSearch, and sliding-window + semantic chunking strategies to retrieve medical guidelines, coverage rules, and care protocols while maintaining HIPAA-compliant access controls.

Engineered NLP extraction pipelines with BioClinicalBERT, spaCy, and custom sequence models to extract ICD, CPT, LOINC, medication, and symptom entities from unstructured clinical notes.

Deployed real-time inference services on AWS SageMaker Endpoints and ECS Fargate, delivering low-latency predictions for anomaly detection, risk scoring, and care-gap identification.

Implemented LLM-driven automation via Amazon Bedrock and prompt-tuned LLaMA/Mistral variants to generate prior-authorization summaries and assess medical-necessity criteria, reducing manual review load during internal testing cycles.

Integrated claims, pharmacy, eligibility, and EHR data using AWS Glue, S3, and PySpark to produce curated, HIPAA-compliant datasets that improved modeling and feature preparation workflows.

Designed clinical Q&A assistants using retrieval-augmented LLMs, few-shot prompting, chain-of-thought patterns, and safety-aligned guardrails to support clinicians with evidence lookup and guideline interpretation.

Implemented MLOps pipelines using SageMaker Pipelines, MLflow, and GitHub Actions for automated retraining, model versioning, drift detection, and CI/CD deployment.

Utilized AWS Lambda for lightweight preprocessing, data validation, and RAG/ML orchestration tasks, and integrated AWS Step Functions to automate clinical review and approval workflows, ensuring structured, auditable execution for GenAI-assisted processes.

Applied Pydantic across FastAPI services and internal ML/GenAI APIs to enforce strong data validation, schema consistency, and type safety for request and response flows.

Authored Terraform modules to provision SageMaker workloads, Fargate services, Glue catalogs, IAM policies, and secure S3 buckets, ensuring repeatable and compliant infrastructure.

Evaluated model and LLM outputs with ROUGE, BLEU, BERTScore, and LangSmith-based tests to monitor summarization stability, retrieval accuracy, and prompt robustness.

Monitored GenAI and ML systems using CloudWatch for logs and alerts, Prometheus for metrics, and Grafana dashboards for real-time visualization of inference performance and service health.

Implemented safety filters, bias checks, PHI redaction, and explainability tooling (SHAP, LIME) to ensure responsible and regulator-aligned deployment of AI systems within clinical workflows.

Environment: Amazon Bedrock, SageMaker, ECS Fargate, AWS Lambda, Step Functions, Glue, S3, EC2, PyTorch, TensorFlow, BioClinicalBERT, LangChain, FAISS, OpenSearch, Pydantic, FastAPI, MLflow, GitHub Actions, Terraform, PySpark, CloudWatch, Prometheus, Grafana, Python, SHAP, LIME.

Bank of America Data Scientist / AI ML Engineer March 2019 - Jan 2022 Charlotte, NC

Roles & Responsibilities:

Engineered fraud detection and anomaly-scoring models using XGBoost, LightGBM, and deep autoencoders to analyze high-velocity card transactions, reducing false positives by 17% and accelerating fraud case closure rates.

Designed real-time ML scoring pipelines using Pub/Sub + Dataflow to process millions of transactions per minute, delivering sub-second fraud alerts for card, ACH, and digital payment channels.

Built NLP pipelines using BERT, spaCy, and TF-IDF to extract entities and categorize SAR narratives, KYC documents, and investigator notes, reducing AML review time by 35% and improving triage accuracy.

Prototyped graph-based fraud features using NetworkX, then productionized relationship analytics in Neo4j (Cypher + GDS), uncovering hidden fraud rings and increasing network-level detection recall by 22%.

Developed credit risk, delinquency, and charge-off prediction models using TensorFlow on AI Platform (Vertex AI), improving borrower risk segmentation and reducing manual underwriting exceptions.

Piloted Vertex AI training, pipelines, and model serving, standardizing MLOps workflows, enabling automated retraining, and cutting model promotion time from days to hours.

Engineered a BigQuery-based feature store to consolidate transaction behavior, device fingerprints, merchant risk attributes, and geo-velocity patterns, reducing feature prep time by 40% for new experiments.

Designed model governance and drift-detection workflows using MLflow, Vertex AI Model Monitoring, and statistical tests, ensuring audit readiness and reducing model degradation incidents.

Containerized ML models with Docker and deployed low latency scoring microservices on GKE, providing horizontally scalable fraud/AML inference capabilities during peak traffic periods.

Built SHAP/LIME-driven explainability dashboards integrated with case management systems, enabling investigators and compliance teams to interpret AI-driven decisions with transparent, regulator-ready evidence.

Environment: Python, scikit-learn, XGBoost, LightGBM, TensorFlow, BERT, PyTorch, BigQuery, GCS, Pub/Sub, Dataflow, Cloud Composer, GKE, Vertex AI, Neo4j (Cypher, GDS), NetworkX, Kafka, Spark, MLflow, SHAP, LIME, Docker, SQL, Git

Safeway Data Scientist August 2017 - Jan 2019 Pleasanton, CA

Roles & Responsibilities:

Developed demand forecasting models using ARIMA, Prophet, and Gradient Boosting to predict weekly SKU-level sales across 1K+ stores, reducing forecast error (MAPE) by 22% and improving replenishment accuracy for high-velocity items.

Engineered inventory optimization models leveraging historical sales, vendor lead times, and seasonality indicators, helping supply chain teams cut stockouts by 18% during promotional cycles.

Built and automated feature engineering pipelines in Python and PySpark on AWS EMR, creating lag features, holiday sensitivity curves, and weather-based demand modifiers that improved model performance across perishables and non-perishables.

Designed price elasticity and promotional uplift models using scikit-learn and XGBoost, identifying ideal discount levels and enabling merchandising teams to increase promo ROI by 12%.

Applied NLP techniques (spaCy, TF-IDF, LDA topic modeling) to categorize product descriptions, clean catalog data, and cluster comparable SKUs, reducing misclassified items in the product hierarchy by 30%.

Built customer segmentation and affinity models using K-Means and hierarchical clustering to identify localized buying patterns, supporting targeted promotions that increased category sales in pilot regions.

Conducted A/B testing frameworks to evaluate store layout changes, promotion strategies, and cross-category bundles, providing statistically backed recommendations that improved basket size by 9%.

Developed early recommendation algorithms using collaborative filtering and association rule mining (Apriori) to surface complementary products, improving digital channel add-to-basket rates.

Deployed model-ready datasets and dashboards using AWS S3, Redshift, and Tableau, enabling business stakeholders to track forecast accuracy, promotion lift, and store-level anomalies in near real time.

Collaborated with supply chain, merchandising, and pricing teams to translate retail challenges into ML solutions, delivering insights that informed product placement, assortment decisions, and regional forecasting strategies.

Environment: Python, Pandas, NumPy, scikit-learn, XGBoost, PySpark, AWS EMR, AWS S3, Redshift, spaCy, TF-IDF, Tableau, SQL, Git

FuGenX Technologies Python Developer Jan 2015 - June 2017 Hyderabad, India

Roles & Responsibilities:

Developed Python-based data processing pipelines using pandas and custom parsers to clean, merge, and transform gameplay telemetry and mobile app logs, improving data readiness and reducing manual prep time by 45%.

Built RESTful APIs in Flask to deliver gameplay insights, engagement metrics, and player behavior stats, reducing reporting turnaround time from days to minutes.

Implemented automated ETL workflows using Python + cron and SQL procedures to ingest user events from MySQL/Oracle into analytics tables, improving daily KPI data availability.

Engineered early ML prototypes (churn prediction, session drop-off detection) using scikit-learn, helping product teams identify retention risks ahead of new game releases.

Created log-analysis scripts to detect crashes, bottlenecks, and latency spikes across mobile and backend systems, reducing average debugging efforts by 30%.

Optimized SQL queries, indexes, and stored procedures across MySQL and SQL Server, improving dashboard and reporting performance.

Designed Python-based data validation checks to detect missing events, corrupt payloads, and anomalies, improving analytics accuracy and reducing reprocessing efforts.

Automated CI/CD workflows using Jenkins and GitLab to streamline deployments for internal Python services and reduce release-time defects.

Environment: Python, Flask, Django, MySQL, SQL Server, Oracle, pandas, NumPy, scikit-learn, cron, REST APIs, Git, GitLab, Jenkins

EDUCATION

Bachelor of Technology in Computer Science GITAM University, Hyderabad

Contact this candidate