AI/ML Engineer - MLOps & Agentic Systems Expert

GURUPRAKASH REDDY AVILALA

AI / ML Engineer · MLOps · Agentic AI Systems

***********************@*****.*** 302-***-**** LinkedIn GitHub: guruprakash30

PROFESSIONAL SUMMARY

AI/ML Engineer with 6 years of experience designing and deploying intelligent systems in healthcare and financial services on Azure. Hands-on expertise across the full ML lifecycle — from feature engineering, model training, and evaluation to cloud deployment, endpoint management, and production monitoring. Experienced integrating LLMs into enterprise workflows through prompt engineering, RAG pipelines, and agentic orchestration with LangGraph, LangChain, and Semantic Kernel; uses Azure Cosmos DB for agent state persistence, LangSmith for observability, and SHAP for model explainability in regulated industries. Bridges the gap between ML development and cloud-native infrastructure — bringing deep knowledge of AKS, CI/CD automation, API security, and distributed systems including Azure Event Hubs and Durable Functions that most ML practitioners lack. Holds Microsoft AI Fundamentals and OCI AI Foundations certifications.

PERSONAL PROJECTS

Automated PR Review Agent — Multi-Agent System with HITL

Source code: github.com/guruprakash30/netflix-recommend-nbs/tree/main/pr_review · Live example PR: github.com/guruprakash30/netflix-recommend-nbs/pull/1

•Built a fully automated GitHub pull request review agent using LangGraph that analyzes code diffs, identifies security vulnerabilities, generates actionable suggestions, and posts structured review comments — without any human intervention for clean PRs.

•Designed a per-file analysis pipeline where a LangChain sub-agent autonomously fetches related source files from the repository, reasons about code dependencies, and produces file-level verdicts (lgtm / needs_changes / security_issue) with token-budget enforcement to stay within LLM context limits.

•Implemented a Human-in-the-Loop (HITL) flow that allows engineers to respond directly in PR comments (approve / reject / chat) — the agent resumes context from its prior review, refines suggestions based on feedback, and can push AI-generated code fix commits directly to the branch.

•Entire review state — diffs, suggestions, security findings, conversation history — persists inside the PR comment itself, enabling stateful multi-turn review sessions with zero backend infrastructure.

Stack: Python · LangGraph · LangChain · Azure OpenAI (GPT-4.1-mini) · Pydantic · GitHub REST API · tiktoken

Netflix-Scale Recommendation System — Dual Model Architecture

Source code: github.com/guruprakash30/netflix-recommend-nbs

Decommissioned endpoints — ACI: 31668e25-5ffc-40f7-b3cf-c7458c0c80b4.eastus.azurecontainer.io/score · Managed Endpoint: cbf-movie-recommender.eastus.inference.ml.azure.com/score

•Built two independent recommendation engines — a collaborative filtering model using Matrix Factorization SVD trained on user-movie rating data, and a content-based filtering model using a HuggingFace Sentence Transformer that matches movies to natural-language queries via semantic similarity — each solving a different cold-start and personalization scenario.

•Trained the collaborative filter with hyperparameter search across matrix rank, learning rate, and regularization using cross-validation; evaluated with both ranking metrics (RMSE) and classification metrics (Precision, Recall, F1) to measure recommendation quality at multiple confidence thresholds.

•Built end-to-end Azure ML pipelines for both models — data ingestion, preprocessing, training, evaluation, artifact registration, and deployment — serving predictions through live Azure ML Managed Endpoints and Azure Container Instance inference APIs.

Stack: Python · PyTorch · HuggingFace Transformers · Scikit-Surprise · Azure ML SDK V1/V2 · Azure Container Registry · Managed Endpoints

ResumeAI — RAG-Based Candidate Screening Pipeline

Source code: github.com/guruprakash30/ResumeAI

•Designed a dual-pipeline RAG system for HR candidate screening — the HR side ingests a job description, uses an LLM to extract and weight required skills, and builds a structured knowledge graph in Neo4j; the candidate side parses uploaded resumes, embeds them using OpenAI embeddings, and indexes them in Azure AI Search as a vector store.

•Implemented hybrid search combining semantic vector similarity (KNN on embedded resume chunks) with keyword search (BM25) in a single retrieval call — improving ranking accuracy over pure vector or keyword approaches by leveraging both semantic context and exact skill term matching.

•At query time, the system traverses the Neo4j graph and the vector index together, ranks candidates by skill-overlap against the job requirements, and synthesizes a structured shortlist using GPT-4 — enabling HR teams to query in natural language (e.g. 'find senior Python engineers with fintech experience') and receive ranked, explainable results.

Stack: .NET 8 ASP.NET Core · Azure AI Search · Azure OpenAI · Neo4j Graph DB · OpenAI Embeddings · Azure Blob Storage · Managed Identity

TECHNICAL SKILLS

Agentic AI & LLMs: LangGraph, LangChain, Semantic Kernel, AutoGen, CrewAI, Azure OpenAI (GPT-4, GPT-4.1-mini), OpenAI Embeddings, RAG, GraphRAG, Prompt Engineering, Function Calling, Structured Output, HITL Workflows, Multi-Agent Orchestration, LangSmith, Azure AI Foundry, Azure AI Search, Azure Content Safety

ML / Deep Learning: PyTorch, HuggingFace Transformers, Scikit-Learn, Scikit-Surprise, XGBoost, LightGBM, Gradient Boosting, Matrix Factorization SVD, Cosine Similarity, Feature Engineering, TF-IDF, Sentence Embeddings, SHAP, LIME, Classification, Regression, Pandas, NumPy, Matplotlib

MLOps & Model Serving: Azure ML SDK V1/V2, Azure ML Pipelines, Managed Endpoints, ACI, Model Registry, MLflow, Experiment Tracking, Dataset Versioning, Azure Container Registry, Docker, Batch Inference, Model Monitoring, Drift Detection, LangSmith, Weights & Biases (W&B)

Vector Stores & Databases: Azure AI Search (KNN + BM25 hybrid), Azure Cosmos DB (NoSQL / conversational state), Redis (Azure Cache for Redis), Neo4j Graph DB (Cypher), Chroma, Qdrant, Pinecone, PostgreSQL, SQL Server, Oracle 11g

Cloud & DevOps: Azure (AKS, Key Vault, Blob Storage, API Management, Service Bus, Event Hubs, Logic Apps, App Insights, Managed Identity, SignalR, Cache for Redis, Durable Functions), Kubernetes, Helm, NGINX, Azure DevOps CI/CD, OCI (OKE, Object Storage)

Backend & APIs: Python, FastAPI, .NET 8 / ASP.NET Core 6/8, C#, REST APIs, Microservices, CQRS, Event-Driven Architecture

CERTIFICATIONS

•Microsoft AI Fundamentals (AI-900)

•Oracle Cloud Infrastructure 2025 AI Foundations Associate

•Oracle Cloud Infrastructure 2025 Certified Architect Associate

•Oracle Cloud Infrastructure 2025 Certified Foundations Associate

•ICPC International Coding Competition

•Facebook Hacker Cup 2022

PROFESSIONAL EXPERIENCE

AI/ML Engineer — Architect · Claritev (Healthcare), Dallas, TX May 2024 – Present

•LLM-powered document intelligence: Integrated Azure OpenAI GPT-4 into healthcare claims workflows to extract structured information from unstructured EOB documents and clinical notes; engineered system prompts for deterministic JSON extraction, implemented token management and retry logic, and layered Azure Content Safety guardrails for HIPAA-compliant PII redaction and output filtering before downstream processing.

•RAG pipeline for clinical knowledge retrieval: Built a Retrieval-Augmented Generation system over internal clinical coding guidelines and policy documents using Azure AI Search (vector + keyword hybrid retrieval); enabled operations teams to query proprietary knowledge bases in natural language and receive accurate, source-cited answers — reducing manual lookup time significantly.

•LLM observability & agent tracing: Instrumented LangChain and LangGraph pipelines with LangSmith for end-to-end trace visibility — capturing prompt inputs, tool calls, token usage, and per-step latency; used trace data to diagnose retrieval failures and prompt regressions in production, enabling systematic prompt iteration and benchmarking across model versions.

•ML model deployment & lifecycle management: Deployed classification models to Azure ML Managed Endpoints with autoscaling; managed model versioning with staged promotion (dev staging prod) in Azure ML Model Registry; configured production monitoring for prediction drift and endpoint latency using Azure Application Insights.

•Feature engineering for healthcare data: Designed Python data pipelines (Pandas, NumPy) processing medical claims datasets for downstream ML models; engineered features for claim duplication detection and coordination-of-benefits overlap prediction; tracked all experiments with Azure ML Run for reproducibility and audit compliance.

•Agentic orchestration for claims triage: Designed multi-step LangChain agent workflows that autonomously query internal APIs, retrieve policy context, and route claims with structured outputs; persisted LangGraph agent graph state and multi-turn conversation history in Azure Cosmos DB using the LangGraph checkpoint pattern, enabling resumable agentic workflows across distributed invocations; implemented human-in-the-loop approval gates for low-confidence decisions, balancing automation throughput with clinical accuracy requirements.

•MLOps & CI/CD: Built Azure DevOps YAML pipelines for automated model retraining, evaluation gating, and AKS deployment; implemented blue/green endpoint traffic splitting for zero-downtime model updates; containerized all ML scoring services with Docker for consistent environment parity across dev and production.

AI/ML Engineer · Capital One (Financial Services), Dallas, TX June 2023 – April 2024

•Fraud & risk classification models: Trained and evaluated binary classification models (Gradient Boosting, XGBoost, Logistic Regression) on credit card transaction datasets for fraud signal detection; used AUC-ROC, Precision-Recall curves, and F1 at multiple operating thresholds; applied SHAP (SHapley Additive exPlanations) to explain feature-level contributions per prediction for regulatory transparency and bias auditing; validated model behavior across customer segments to prevent disparate impact in credit decisioning.

•Feature engineering & feature store: Built reusable feature pipelines in Python — engineered transaction velocity features, rolling window aggregates, merchant category encodings, and time-based behavioral signals; versioned feature sets in Azure ML Datasets to ensure consistent feature availability across training and serving environments.

•Real-time model serving at scale: Deployed Scikit-Learn models as FastAPI REST services containerized with Docker on Azure Kubernetes Service; sourced real-time transaction events from Azure Event Hubs for streaming inference, enabling sub-second fraud signal generation on incoming payment streams; integrated inference endpoints into ASP.NET Core backends serving real-time credit decisioning for 2M+ active users; applied horizontal pod autoscaling to handle peak transaction volumes.

•NLP for customer interaction analysis: Applied NLP techniques including TF-IDF and sentence embeddings to customer service transcripts for intent classification and escalation routing; evaluated with macro-averaged F1 across multiple intent classes; integrated model predictions into case management workflows to reduce manual triage.

•Experiment tracking & model governance: Managed hyperparameter tuning runs and artifact versioning in Azure ML with MLflow integration — tracking metrics, parameters, and model artifacts across training iterations for reproducible comparisons; built A/B evaluation framework to compare challenger vs. champion models before production promotion; maintained full experiment lineage and metric logging for regulatory audit requirements in financial services.

•Data drift monitoring & retraining: Implemented statistical drift detection (KS test on input feature distributions) as a scheduled Azure ML pipeline step; configured Application Insights alerting on prediction score distribution shifts; defined retraining trigger thresholds and automated retraining pipeline execution on drift detection.

.NET Developer · Temenos, Chennai, India January 2020 – December 2021

•AI/ML integration exposure: Contributed to early-stage integration of energy consumption anomaly detection logic using statistical thresholds and rule-based pattern detection on time-series telemetry data; worked alongside data engineers to expose model predictions through RESTful APIs consumed by real-time dashboards.

•Data pipeline development: Built batch processing pipelines reading flat files and XML to generate structured energy consumption reports; implemented stored procedures, triggers, and SSIS-based ETL workflows for data migration between on-premises databases and Azure SQL Database.

•Cloud deployment & DevOps: Deployed ASP.NET MVC Core and Angular 11 applications to Azure App Service with Azure DevOps CI/CD; configured Azure Application Insights for distributed tracing and Azure AD B2C with OAuth 2.0 / OpenID Connect for API authentication.

EDUCATION

Master of Science in Information Technology · University of Cincinnati, US · Jan 2022 – Apr 2023 · GPA: 4.0/4.0

Contact this candidate