AI Engineer

VIKYATH REDDY

SENIOR AI/ML & GENERATIVE AI ENGINEER AGENTIC AI MLOPS

857-***-**** **************@*****.*** Linkedin

PROFESSIONAL SUMMARY

•Senior AI/ML and Generative AI Engineer with 8 years of progressive experience building production AI, machine learning, data and cloud-native systems across banking, insurance, enterprise data management, medical technology and healthcare.

•Career progression from healthcare data science and predictive modeling to production deep learning, enterprise ML product engineering, insurance GenAI, and current agentic AI and enterprise RAG engineering in banking.

•Strong production Python engineering background spanning FastAPI, REST APIs, asynchronous processing, model and agent serving, automation, testing, CI/CD and cloud-native deployment.

•Hands-on Agentic AI experience using LangGraph, LangChain, Model Context Protocol (MCP), AutoGen and CrewAI patterns for multi-step workflows, governed tool/function calling, stateful orchestration, enterprise API integration and human-in-the-loop controls.

•Extensive Retrieval-Augmented Generation experience covering ingestion, structure-aware chunking, embeddings, hybrid semantic and keyword retrieval, metadata filtering, query rewriting, reranking, grounding, citations and retrieval-quality evaluation.

•Strong machine-learning and deep-learning foundation using Scikit-learn, XGBoost, PyTorch, TensorFlow and Keras for classification, fraud/risk scoring, NLP, entity matching, recommendation, forecasting, anomaly detection and high-volume sensor analytics.

•Built real-time and batch AI/data pipelines using Databricks, Spark/PySpark, Spark Structured Streaming, Delta Lake, Kafka, Event Hubs, Kinesis, Airflow, Snowflake and BigQuery, with emphasis on reliable model-ready data and low-latency features.

•Production MLOps and LLMOps experience with MLflow, SageMaker, Vertex AI, model registries, CI/CD, automated evaluation, staged promotion, rollback, drift monitoring, retraining and model/agent observability.

•Built and operated cloud-native AI services with Docker and Kubernetes across AKS, EKS and GKE, using Terraform, Azure DevOps, GitHub Actions and Jenkins for repeatable infrastructure and application delivery.

•Multi-cloud experience tied to real project contexts: Azure-centered banking AI at PNC, AWS-centered claims AI at The Hartford, GCP/Vertex AI enterprise ML at Informatica, and AWS-based medical ML workloads at Medtronic.

•Experienced implementing security and governance controls including OAuth/RBAC/IAM, secrets handling, PII protection, prompt-injection defenses, content-safety checks, audit logging, traceability and human approval for sensitive AI actions.

•Comfortable translating business and domain requirements into technical designs, working across product, data, platform, risk/compliance and business teams through architecture reviews and release planning.

TECHNICAL SKILLS

Programming & Backend: Python, SQL, PySpark, Go, TypeScript/JavaScript, Java, FastAPI, REST APIs, Microservices, Async Python, Pydantic, Pandas, NumPy

Machine Learning & Deep Learning: Scikit-learn, XGBoost, PyTorch, TensorFlow, Keras, NLP, Classification, Regression, Anomaly Detection, Time Series, Recommendation, Feature Engineering, Model Evaluation

Generative AI & Agents: Azure OpenAI, Amazon Bedrock/Claude, LangChain, LangGraph, MCP, AutoGen, CrewAI, RAG, Tool/Function Calling, Prompt Engineering, Structured Outputs, Multi-Step Workflows, Human-in-the-Loop

LLM Retrieval & Search: Embeddings, Semantic/Hybrid Search, Metadata Filtering, Query Rewriting, Reranking, Grounding, Citations, Azure AI Search, OpenSearch, Pinecone, FAISS, ChromaDB

Data Engineering: Databricks, Apache Spark, Spark Structured Streaming, Delta Lake, Medallion Architecture, Unity Catalog, Azure Data Factory, Kafka, Event Hubs, Kinesis, Snowflake, BigQuery, dbt, Airflow, Data Quality

MLOps / Cloud: MLflow, SageMaker, Vertex AI, Docker, Kubernetes, AKS, EKS, GKE, Terraform, Azure DevOps, GitHub Actions, Jenkins, CI/CD, Model Registry, Drift Monitoring

Security / Observability: OAuth, RBAC/IAM, Managed Identities, PII Controls, Audit Logging, Responsible AI, OpenTelemetry, Prometheus, Grafana, Azure Monitor, CloudWatch

Databases / Analytics: PostgreSQL, Redis, Cosmos DB, MongoDB, DynamoDB, SQL Server, Power BI

EDUCATION

Bachelor of Technology in Computer Science and Engineering

Vardhaman College Of Engineering, Hyderabad, Telangana, 2017

PROFESSIONAL EXPERIENCE

PNC Bank, Pittsburgh, PA (Sept 2024 - Present)

Senior AI/ML Engineer Python, Azure OpenAI, LangGraph, MCP, Databricks, Kubernetes

Project / Work Summary: Lead production AI/ML engineering for banking fraud, AML, enterprise knowledge and operational-assistance workflows, combining real-time ML/data engineering, enterprise RAG, governed agentic AI and MLOps on Azure.

•Design and build agentic GenAI applications for retail and commercial banking operations on Azure, using LangGraph to orchestrate multi step agent workflows for dispute investigation, KYC document review, and banker facing research assistants, cutting average case handling time by 35%.

•Develop Model Context Protocol (MCP) servers in Python that expose core banking capabilities such as customer profile lookup, transaction history retrieval, and case management actions as governed tools for LLM agents, with OAuth based authentication, role based access, and full audit logging, supporting 15+ governed tools in production.

•Build a real time fraud detection pipeline that scores card and Zelle transactions in flight, consuming events from Azure Event Hubs into Spark Structured Streaming on Databricks and serving gradient boosted and deep learning models with sub second latency through an online feature store, reducing fraud losses by 22% year over year.

•Implement real time AML transaction monitoring enrichment, streaming alerts into an LLM powered triage layer that summarizes customer activity, correlates related alerts, and drafts investigator narratives, cutting manual case preparation effort for the financial crimes team by 40%.

•Engineer an enterprise RAG platform over banking policies, compliance manuals, and product documentation using Azure AI Search with hybrid vector plus keyword retrieval, semantic reranking, and grounded answers with citations for branch and contact center staff, reducing average query resolution time by 30%.

•Fine tune and evaluate GPT-4o and GPT-4.1 deployments on Azure OpenAI for banking specific summarization and classification tasks, applying prompt versioning, structured outputs with function calling, and token and cost optimization across environments, cutting per query inference cost by 25%.

•Establish an LLM evaluation harness with golden datasets, RAGAS style faithfulness and answer relevance metrics, and LLM as judge scoring wired into CI pipelines so every prompt, retrieval, or model change is regression tested before release, improving retrieval precision by 18 percentage points.

•Apply guardrails including Azure AI Content Safety, PII detection and redaction with Presidio, prompt injection defenses, and human in the loop approval steps for any agent action that modifies customer or account data.

•Build and operate FastAPI microservices for model and agent serving, packaged with Docker and deployed to Azure Kubernetes Service with Helm charts, horizontal pod autoscaling, and blue green releases through Azure DevOps pipelines, supporting 10+ services at 99.9% uptime.

•Manage the ML lifecycle on Databricks with MLflow for experiment tracking, model registry, and staged promotion, plus Delta Lake feature tables shared between streaming inference and batch training to prevent training serving skew.

•Monitor production models and agents with Azure Monitor, Prometheus, and Grafana dashboards tracking latency, drift, hallucination flags, and guardrail triggers, and run scheduled retraining when data drift thresholds are breached, cutting mean time to detect model or agent issues by 45%.

•Partner with model risk management, information security, and compliance teams to document model cards, data lineage, and validation evidence for GenAI use cases, aligning delivery with bank model governance requirements across 6+ audit cycles.

Environment: Python, Azure OpenAI (GPT-4o, GPT-4.1), LangGraph, LangChain, MCP, Azure AI Search, Databricks, Delta Lake, Spark Structured Streaming, Azure Event Hubs, Kafka, MLflow, FastAPI, Docker, AKS, Helm, Azure DevOps, Terraform, Cosmos DB, PostgreSQL, Redis, Presidio, Azure AI Content Safety, Prometheus, Grafana

The Hartford, Hartford, CT (Feb 2023 - Jul 2024)

Senior Machine Learning Engineer Python, AWS SageMaker, Amazon Bedrock, LangChain, Kinesis

Project / Work Summary: Expanded from traditional ML into cloud-scale claims AI and Generative AI, building real-time predictive services, document intelligence and an adjuster-facing RAG assistant on AWS with evaluation and guardrails.

and Lambda; Bedrock (Claude) claims assistant with Textract document intelligence and OpenSearch retrieval.

•Built a real time First Notice of Loss (FNOL) triage system that ingested claim submissions through Amazon Kinesis Data Streams and scored severity and complexity within seconds, routing high severity auto and property claims to senior adjusters automatically, cutting triage time from 30 to 8 minutes.

•Deployed real time inference endpoints on Amazon SageMaker with autoscaling and multi model endpoints, serving severity, fraud propensity, and litigation likelihood models to claims intake systems over low latency REST APIs.

•Developed a GenAI claims assistant on Amazon Bedrock using Claude models with LangChain, letting adjusters ask natural language questions over claim files, policy documents, and adjuster notes with retrieval from Amazon OpenSearch vector indexes, cutting adjuster research time by 35%.

•Engineered document intelligence pipelines with Amazon Textract and Hugging Face transformer models to extract entities, coverage details, and damage descriptions from ACORD forms, police reports, and medical records feeding downstream claim models.

•Implemented LLM powered claim file summarization that condensed lengthy adjuster notes and correspondence into structured summaries at key claim lifecycle events, reducing time adjusters spent reconstructing claim history by 40%.

•Processed streaming telematics style driving and IoT signals with Kinesis and AWS Lambda to compute near real time risk features supporting usage based auto insurance analytics and pricing experimentation.

•Designed feature engineering and training pipelines with SageMaker Processing and SageMaker Pipelines over data in Amazon S3 and Snowflake, with SageMaker Feature Store providing consistent online and offline features for real time scoring.

•Fine tuned transformer models for insurance specific text classification tasks such as claim cause coding and subrogation opportunity detection, benchmarking against XGBoost baselines and shipping the best performer per line of business, improving classification accuracy by 12 points.

•Set up evaluation and guardrails for GenAI features, combining grounding checks, PII masking before prompts left the VPC, prompt injection testing, and human review queues for low confidence outputs in regulated claim decisions.

•Automated CI/CD for ML services with GitHub Actions and Terraform, packaging FastAPI scoring services into Docker containers deployed on Amazon EKS with canary rollouts and CloudWatch alarms on latency and error budgets, cutting deployment time from 2 days to 3 hours.

•Monitored model health with SageMaker Model Monitor and custom drift jobs, tracking population stability of key features and triggering retraining workflows when claim mix shifted after catastrophe events, catching 90% of drift before it affected claim decisions.

•Collaborated with claims operations, actuarial, and data governance partners to validate model behavior, document assumptions, and align GenAI usage with company AI risk and compliance standards.

Environment: Python, AWS (SageMaker, Bedrock, Kinesis, Lambda, EKS, S3, Textract, OpenSearch, CloudWatch, IAM), LangChain, Hugging Face Transformers, PyTorch, XGBoost, FastAPI, Docker, Terraform, GitHub Actions, Airflow, Snowflake, PostgreSQL, Redis

Informatica, Redwood City, CA (Nov 2021 - Jan 2023)

Machine Learning Engineer Python, GCP, Machine Learning, NLP, Metadata Intelligence, MLOps

Project / Work Summary: Built ML/NLP capabilities for enterprise data-management products, moving models from experimentation into reusable services for metadata classification, semantic discovery, entity matching and data-quality intelligence on GCP.

•Built machine learning models powering the platform's CLAIRE AI engine capabilities for a leading enterprise data management product, automating data classification, domain detection, and PII identification across large customer data estates spanning 50M+ records.

•Developed NLP models to infer semantic data types and business terms from column names, sample values, and metadata, improving automated data cataloging and cutting manual tagging effort for enterprise customers by 30%.

•Built ML based entity matching and record deduplication models for master data management, using fuzzy matching, embeddings, and similarity scoring to resolve duplicate customer and product records at scale, improving match precision by 20%.

•Engineered anomaly detection models over data quality metrics and pipeline telemetry, flagging schema drift, volume anomalies, and freshness issues before they affected downstream analytics, cutting data incidents by 25%.

•Designed embedding based similarity and search over metadata and data assets, helping users discover related datasets across a large enterprise data catalog.

•Trained and evaluated models on Google Cloud with Vertex AI, using BigQuery for large scale feature preparation and Cloud Storage for datasets and artifacts, with reproducible pipelines and experiment tracking in MLflow.

•Built recommendation models that suggested data transformations, mappings, and data quality rules to users, learning from historical mapping patterns across projects.

•Packaged models as FastAPI microservices, containerized with Docker and deployed on Google Kubernetes Engine with autoscaling and CI/CD through GitHub Actions.

•Prototyped early Generative AI features for natural language data discovery, letting users describe the data they needed in plain language and retrieving matching assets through semantic search.

•Partnered with product and platform engineering to move models from research into the cloud data management product, defining APIs, latency targets, and monitoring.

•Established model monitoring for classification accuracy and drift on live customer metadata, triggering retraining as new data domains and patterns emerged.

Environment: Python, GCP (Vertex AI, BigQuery, GKE, Cloud Storage), Machine Learning, NLP, Hugging Face Transformers, PyTorch, Scikit-learn, Embeddings, Entity Resolution, MLflow, FastAPI, Docker, GitHub Actions, Terraform, Airflow

Medtronic, Minneapolis, MN (Feb 2020 - Sep 2021)

Machine Learning Engineer Python, TensorFlow, AWS SageMaker, Spark, Time Series

Project / Work Summary: Built production ML and deep-learning workflows for medical-device telemetry, progressing into large-scale data preparation, time-series modeling, cloud training and model validation in a regulated environment.

•Developed machine learning models on continuous glucose monitoring and insulin pump telemetry, building time series forecasting and classification models that supported glycemic pattern insights for diabetes management analytics.

•Built deep learning models in TensorFlow and Keras for physiological signal processing, including CNN and LSTM architectures for waveform pattern detection and event classification on high frequency device data.

•Engineered anomaly detection pipelines over streaming device telemetry to flag sensor faults and abnormal readings, combining statistical process control baselines with autoencoder based detectors, cutting false sensor fault alerts by 30%.

•Trained and tuned models on Amazon SageMaker with managed training jobs, hyperparameter optimization, and spot instances, storing versioned datasets and artifacts in Amazon S3 with strict access controls for patient related data.

•Deployed inference services through SageMaker endpoints and AWS Lambda for batch and near real time scoring, integrating outputs into analytics platforms used by clinical and product teams.

•Built large scale data preparation pipelines with PySpark on Amazon EMR, cleaning, resampling, and windowing billions of device readings into model ready training datasets in S3 based data lakes.

•Implemented rigorous model validation protocols including stratified cross validation, sensitivity/specificity analysis, and subgroup performance review to meet regulated medical device design-control standards.

•Collaborated with clinical scientists and regulatory partners to document data provenance, model assumptions, and verification evidence supporting internal review of algorithm changes.

•Developed reproducible experiment workflows with MLflow tracking, Git based code review, and containerized training environments in Docker to ensure consistent results across teams.

•Created monitoring dashboards for deployed models covering input data quality, prediction distributions, and drift, with CloudWatch alarms alerting on data pipeline failures.

•Optimized AWS costs for training and inference by right sizing instances, adopting spot capacity for experimentation, and archiving cold telemetry data to lower cost S3 tiers, cutting infrastructure costs by 20%.

Environment: Python, TensorFlow, Keras, Scikit-learn, XGBoost, PySpark, AWS (SageMaker, EMR, S3, Lambda, CloudWatch, IAM), MLflow, Docker, Airflow, PostgreSQL, Git, Jira

Fortis Healthcare, Gurugram, India (Feb 2017 - Dec 2019)

Data Scientist Python, SQL, Scikit-learn, NLP, Azure ML

Project / Work Summary: Established the core data-science foundation through healthcare predictive modeling, SQL/Python data preparation, clinical NLP, statistical analysis and model validation for hospital operations and preventive-care use cases.

•Built predictive models for patient readmission risk and length of stay using Scikit-learn and XGBoost on hospital information system data, supporting care coordination teams in prioritizing post discharge follow ups, contributing to a 15% reduction in 30 day readmissions.

•Developed cardiovascular risk scoring models on longitudinal patient health data, supporting the hospital's preventive cardiology analytics program on Azure.

•Applied NLP to unstructured clinical notes and discharge summaries, using TF-IDF features and classical classifiers to extract diagnoses mentions and support clinical coding.

•Trained and evaluated models on Azure Machine Learning Studio, using its pipelines for feature selection, algorithm comparison, and publishing predictive web services for internal applications.

•Designed SQL based data extraction and cleansing routines over EMR and lab systems, resolving inconsistent codings and missing values for reliable analytical datasets.

•Performed statistical analysis and hypothesis testing on treatment outcome and operational datasets, presenting findings on bed utilization and diagnostic turnaround times to hospital administrators.

•Built patient segmentation models with clustering techniques to support targeted health check package recommendations and chronic disease outreach programs.

•Created recurring dashboards and reports in Power BI and Excel for clinical quality indicators, infection rates, and department level KPIs consumed by medical and operations leadership.

•Automated recurring data preparation and scoring jobs with Python scripts scheduled on Azure virtual machines, reducing manual reporting effort across analytics teams by 35%.

•Worked closely with physicians, nursing informatics, and IT teams to translate clinical questions into analytical problems and validate outputs against clinical judgment.

Environment: Python, Scikit-learn, XGBoost, Pandas, NumPy, NLTK, SQL Server, Azure Machine Learning Studio, Azure Virtual Machines, Power BI, Excel, Git

Contact this candidate