Home

Shiva Velichalamala - AI & ML Engineer

[email protected]

Location: Seattle, Washington, USA

Relocation: Yes

Visa:

Resume file: Shiva_velichamala_AI & Ml Engineer resume_1765205694318.pdf Please check the file(s) for viruses. Files are checked manually and then made available for download.

Shiva Velichalamala Senior AI & ML Engineer Email: [email protected] | Phone: +1 (614) 706-7619| LinkedIn PROFESSIONAL SUMMARY AI/ML Engineer and a Senior Data Scientist with 9+ years of experience designing, deploying, and optimizing AI-driven workflow automation systems across finance, health, and insurance domains. Focused on building LLM-based RAG and agentic AI systems, integrating GPT-4, LangChain, and LangGraph to enable enterprise-scale automation, intelligent decision-making, and workflow optimization. Proficient in Python with expertise across PyTorch, TensorFlow, Keras, and modern frameworks (FastAPI, Flask, Django). Strong background in data science & analytics, leveraging Pandas, NumPy, Scikit-learn, Dask, Spark, and BigQuery to perform advanced data wrangling, feature engineering, and large-scale distributed processing. Hands-on experience with LangGraph, Model Context Protocol (MCP), Guardrails AI and multi-agent coordination frameworks. Strong foundation in data-centric AI building, optimizing, and monitoring data pipelines with Spark, Airflow, and Databricks to ensure model reliability and traceability. Hands-on with model evaluation and governance using SHAP, LIME, Guardrails AI, and Evidently AI to ensure explainability, bias detection, and production transparency. Expert in Generative AI and multimodal systems, leveraging text, image, and audio models (Whisper, Stable Diffusion) for contextual understanding and automation. Proven success integrating MLOps practices (CI/CD, Docker, Kubernetes, SageMaker) to deliver secure, reproducible, and scalable AI deployments. Delivered measurable impact, 50% faster loan approvals, 30% improvement in insight accuracy, and 15% reduction in financial risk across enterprise AI projects. Experienced in building end-to-end data pipelines, predictive models, and real-time analytics workflows that deliver actionable insights and support enterprise decision-making. Expert in LLM-based intelligent automation, LangGraph-orchestrated workflows, and RAG pipelines using GPT-4, LangChain, Pinecone, and Weaviate to drive FinTech back-office efficiency and regulatory compliance. Skilled in cloud & DevOps, with expertise in AWS SageMaker, Lambda, ECS, Azure Cognitive Services, Docker, Kubernetes, Helm, and Terraform to dealiver scalable, secure, and production-ready AI/ML deployments. Expert in relational and NoSQL databases including PostgreSQL, MySQL, MongoDB, Redis, and Elasticsearch, with experience in query optimization, data modeling, and high-performance pipelines. Expertise in Generative AI models (GPT-3/4/5, LLaMA 2, Stable Diffusion, DALL E, Whisper) and NLP frameworks (Hugging Face Transformers, spaCy, NLTK) for developing advanced conversational and text-generation solutions. Designed and implemented RAG pipelines using LangChain, Pinecone, FAISS, and Elastic Search to enhance retrieval efficiency, achieving a 30% improvement in accuracy of financial insights through optimized information access. Proven ability to deploy production-grade AI models and backend Python microservices (FastAPI, Flask) integrated with Airflow, Spark, and Kafka pipelines, delivering compliant automation for loan and risk workflows. Developed multimodal AI solutions combining text, image, and speech processing for advanced real-world applications, utilizing Stable Diffusion, Whisper, and custom LLMs to deliver seamless cross-modal intelligence. Fine-tuned Large Language Models (LLMs) for domain-specific use cases across healthcare, finance, and retail, aligning with compliance standards to improve accuracy, contextual understanding, and production-grade performance. Designed and implemented Agent-to-Agent (A2A) communication protocols enabling autonomous collaboration, dynamic task delegation, and seamless coordination among LLM-based agents. Expertise in cloud-native AI deployments using AWS SageMaker, Azure Cognitive Services, and GCP Vertex AI, integrating CI/CD automation pipelines for secure, scalable, and efficient production workflows. Implemented MLOps practices including CI/CD (GitHub Actions, Jenkins, GitLab CI), containerization (Docker, Kubernetes, Helm), and experiment tracking (MLflow, Weights & Biases). Delivered predictive analytics and anomaly detection systems that significantly reduced fraud and financial risks. Developed interactive dashboards and visualization tools with Tableau, Power BI, Matplotlib, Plotly, and Seaborn to track AI performance, enabling real-time monitoring and data-driven decision-making. TECHNICAL SKILLS Programming Languages Python (Advanced), SQL, Scala, JavaScript (Basic), Shell Scripting AI / Machine Learning Frameworks TensorFlow, PyTorch, Keras, Scikit-learn, XGBoost, LightGBM, CatBoost, ONNX, Hugging Face Transformers Generative AI & LLM Ecosystem GPT-3/4/5, Claude, Gemini, LLaMA-2, Falcon, Mistral, LangChain, LangGraph, Model Context Protocol (MCP), Guardrails AI, PromptLayer, LoRA / QLoRA, PEFT, LangSmith, OpenAI API, Vertex AI Agent Builder, AWS Bedrock. RAG & Vector Databases Pinecone, FAISS, Weaviate, ChromaDB, ElasticSearch, Vector Embeddings, Retrieval Evaluation (BLEU, ROUGE, F1) Data Science & Analytics Pandas, NumPy, SciPy, Dask, Spark MLlib, Databricks, Delta Lake, Feature Engineering, SHAP, LIME, Evidently AI (Drift Detection), Time-Series Forecasting (Prophet), Clustering, PCA Cloud & MLOps Platforms AWS (SageMaker, Bedrock, Lambda, ECS, S3, Redshift), Azure (AI Studio, Cognitive Services, Data Factory, Synapse, AKS), GCP (Vertex AI, BigQuery), MLflow, Weights & Biases, Kubeflow, CI/CD (GitHub Actions, Jenkins, GitLab CI), Docker, Kubernetes, Helm, Terraform Data Engineering & Pipelines Apache Spark, PySpark, Airflow, dbt, Kafka, Delta Lake, ETL/ELT Design, Data Validation, Feature Store (Feast) Visualization & Monitoring Tableau, Power BI, Plotly, Matplotlib, Seaborn, Grafana, Prometheus, Streamlit, Dash Databases Relational (PostgreSQL, MySQL) and Non-relational (MongoDB, Elasticsearch, Redis) Snowflake, Hive, Oracle 11g. Security & Privacy HIPAA, SOC2, GDPR compliance in AI/ML workflows Big-Data Framework Hadoop Ecosystem 2.X (HDFS, MapReduce, Hbase 0.9), Spark Framework 2.X (Scala 2.X, Spark SQL, Pyspark, Spark, Mllib) AI Infrastructure and Platforms NVIDIA CUDA (A100, T4), TensorRT, Mixed-Precision Training, Model Quantization, GPU Workload Profiling PROFESSIONAL EXPERIENCE Client: Alaska Airlines, Seattle, WA Role: Senior AI-ML Engineer Jan 2025 - Present Responsibilities: Delivered a cloud-native ML automation platform using Python, PyTorch, TensorFlow, Spark, and Azure ML, improving model throughput by ~40% and enabling near real-time document intelligence across high-volume clinical and operational document workflows for multiple enterprise teams. Designed an enterprise-grade microservices and event-driven architecture with containerized ML services on AKS, integrating vector indexing, feature stores, and data lake storage on Azure Data Lake and Blob Storage to support scalable training, inference, and analytics workloads. Worked in Agile/Scrum with bi-weekly sprints, collaborating closely with Product Owners, solution architects, data scientists, and platform engineers to refine user stories, groom backlogs, and align ML initiatives with regulatory, operational, and business objectives. Built robust data ingestion pipelines to pull structured and unstructured data from REST APIs, SQL databases, Blob/S3 storage, and Kafka topics, standardizing schemas and metadata for downstream training, feature generation, and inference services. Developed scalable data processing workflows using Pandas, PySpark, and Airflow, performing cleaning, normalization, enrichment, and feature engineering to transform raw clinical and financial documents into ML-ready datasets with traceable lineage. Managed distributed data storage across Azure Data Lake, curated Delta-style layers, and Snowflake warehouses, enabling low-latency access patterns for ML training, analytics queries, and near-real-time scoring workflows. Implemented specialized vector storage solutions using FAISS and Pinecone, supporting large-scale embedding generation, nearest-neighbor search, and semantic retrieval for document similarity, classification, and content-based recommendations. Selected, trained, and productionized ML models using PyTorch, TensorFlow, and scikit-learn for document classification, entity extraction, and risk-related scoring, aligning model architectures with latency, accuracy, and interpretability requirements. Applied advanced techniques such as retrieval-based scoring, vector similarity search, mini-batch inference, and streaming-style pipelines to handle high-volume workloads while maintaining predictable response times and resource utilization. Optimized ML pipelines with hyperparameter tuning, model quantization, caching, and dynamic batching strategies, improving end-to-end inference latency and reducing GPU/CPU costs across production environments. Used frameworks including FastAPI, Spark MLlib, and Hugging Face components to wrap models as reusable services, exposing standardized REST endpoints for internal consumers and downstream orchestration layers. Enforced strong coding standards through modular Python, OOP-based ML components, shared utility libraries, and code review practices, reducing technical debt and simplifying onboarding for new engineers and data scientists. Built comprehensive evaluation workflows with offline validation, cross-validation, drift analysis, and controlled PoC benchmarking, ensuring only rigorously tested models progressed into higher environments and production systems. Containerized all ML workloads using Docker and published images to Azure Container Registry, standardizing runtime environments and enabling consistent deployments across development, staging, and production clusters. Deployed and managed high-availability inference endpoints on AKS using KServe, horizontal autoscaling, and health probes, ensuring reliable performance under fluctuating traffic patterns and seasonal peaks. Implemented automated CI/CD pipelines in GitHub Actions for building, testing, containerizing, and deploying ML models, data pipelines, and microservices with minimal manual intervention and traceable change histories. Used Terraform and Helm to define and provision AKS clusters, networking, storage classes, secrets, and supporting cloud resources as code, ensuring reproducibility, auditability, and consistent infrastructure configurations. Established end-to-end observability with Prometheus, Grafana, and Evidently AI, monitoring latency, resource utilization, data drift, and prediction quality through custom dashboards and automated alerts. Implemented automated unit tests, integration tests, and pipeline validation checks using PyTest and Airflow test hooks, catching data and logic issues early before impacting downstream consumers and production workloads. Authored detailed technical documentation, architecture diagrams, API specifications, and internal runbooks, and led recurring knowledge transfer sessions for platform engineers, data scientists, and operations teams to ensure sustainable ownership. Environment: Python, PyTorch, TensorFlow, scikit-learn, Spark, PySpark, Azure ML, Azure Data Lake, Azure Blob Storage, Snowflake, FAISS, Pinecone, FastAPI, Airflow, Docker, AKS, KServe, MLflow, DVC, Terraform, Helm, Prometheus, Grafana, GitHub Actions Client: State of CA, SFO, CA Role: Senior ML Engineer Aug 2023 Dec 2024 Project: Real-Time Financial Risk Monitoring & Predictive Analytics Platform Responsibilities: Delivered an end-to-end financial risk ML platform using PyTorch, TensorFlow, scikit-learn, and Azure ML, improving risk detection accuracy by ~35% and enabling earlier identification of anomalies across large-scale transactional data streams. Designed a modular ML architecture with containerized microservices, event-driven retraining workflows, centralized model registry, and integrated feature store, supporting governed model promotion and consistent features across multiple risk applications. Participated in Agile/Scrum sprints, collaborating with risk, compliance, data, and product teams to prioritize ML features, refine acceptance criteria, and align model behavior with regulatory and business expectations. Built resilient ingestion pipelines consuming SQL warehouse tables, Azure Blob Storage, REST APIs, and Kafka streams, normalizing disparate financial feeds into unified, versioned ML datasets. Engineered reusable preprocessing and feature pipelines using Python, Pandas, scikit-learn, and PySpark, ensuring consistent transformations between offline training workflows and online inference services. Managed curated ML data layers in Azure Data Lake, Delta-style tables, and Snowflake, tracked through DVC and MLflow for full lineage, experiment repeatability, and audit-ready history. Implemented and maintained a feature store using Feast, enabling centralized definition, low-latency serving, and cross-team reuse of critical risk and behavioral features. Developed ML models for anomaly detection, fraud scoring, and time-series forecasting, targeting transactional irregularities, late payments, and emerging risk patterns in financial portfolios. Applied advanced methods including ensemble modeling, sliding-window forecasting, rare-event modeling, and custom loss functions to improve performance on highly imbalanced datasets with limited labeled positives. Optimized models through systematic hyperparameter tuning, model pruning, ONNX export, and inference graph optimizations, reducing latency and compute costs while preserving required accuracy thresholds. Leveraged PyTorch Lightning, TensorFlow, Spark ML, and FastAPI to standardize training pipelines and expose production-ready inference endpoints integrated with upstream and downstream risk systems. Established robust coding standards, including modular architecture, configuration-driven pipelines, shared utility layers, and code reviews, increasing maintainability and reducing regression risk across ML repositories. Built reusable evaluation frameworks using cross-validation, ROC/AUC, precision/recall, KS statistics, and challenger-versus-champion comparisons to support regulatory model validation and governance. Containerized ML services with Docker and published images to Azure Container Registry, standardizing runtime environments and simplifying promotion between dev, test, and production clusters. Deployed models onto AKS using GPU-backed nodes, HPA-based autoscaling, and blue/green rollout strategies, minimizing downtime and enabling safe, incremental model upgrades. Implemented end-to-end CI/CD using GitHub Actions, Azure DevOps, and Argo Workflows for continuous training (CT) and continuous deployment (CD) of ML models, pipelines, and infrastructure components. Provisioned and managed cloud infrastructure via Terraform, covering AKS clusters, virtual networks, storage accounts, secrets management, and supporting PaaS resources with environment-specific configurations. Set up comprehensive monitoring using Azure Monitor, Prometheus, and Grafana, tracking pipeline health, data drift, model performance, and infrastructure utilization through dashboards and alert rules. Built automated unit tests, integration tests, and UAT support scripts leveraging PyTest and custom validation utilities, ensuring that new model versions and pipeline changes met reliability and compliance standards. Documented end-to-end ML workflows, feature dictionaries, deployment runbooks, and troubleshooting guides, and led onboarding workshops to enable analysts, engineers, and risk stakeholders to adopt and extend the ML platform. Environment: Python, PyTorch, TensorFlow, scikit-learn, XGBoost, Spark, Feast, MLflow, DVC, Airflow, FastAPI, Docker, Kubernetes/AKS, Helm, Terraform, Azure ML, Azure Data Lake, Azure Blob Storage, Azure Monitor, Azure DevOps, Prometheus, Grafana, GitHub Actions Client: Elevance Health, Indianapolis, IN. Role: ML Engineer Mar 2021 Jul 2023 Project: Enterprise Predictive Analytics & Operational Risk Intelligence Platform Responsibilities: Delivered an enterprise-wide ML platform for predictive analytics and operational risk using PyTorch, TensorFlow, XGBoost, and Azure ML, improving risk signal detection and decision-making speed. Designed a modular ML architecture with containerized microservices, a central model registry (MLflow), and batch/real-time scoring flows on AKS. Worked in Agile/Scrum with 2-week sprints, collaborating with Product Owners, data scientists, and engineers to refine ML use cases and acceptance criteria. Built ingestion pipelines using Azure Functions, REST APIs, Azure Blob Storage, and database connectors to onboard financial and operational data. Engineered preprocessing and feature pipelines using Python, Pandas, scikit-learn, and Azure ML Pipelines to standardize transformations across training and inference. Managed storage across Azure Data Lake and curated Delta-style layers feeding analytics and ML workloads. Implemented data quality checks, schema validation, and drift-aware input monitoring to protect model stability in production. Trained models (XGBoost, neural networks in PyTorch/TensorFlow) for risk scoring, anomaly detection, and operational forecasting. Applied ensemble modeling, time-series analysis, and imbalance handling to improve recall on rare risk events. Optimized performance via hyperparameter tuning, cross-validation, and model compression techniques. Used TensorFlow Serving, TorchServe, FastAPI, and MLflow to package, register, and deploy ML models at scale. Enforced modular Python code, reusable ML components, and configuration-driven pipelines with rigorous code reviews. Built evaluation workflows with MLflow experiments and standardized metrics to select and promote models to production. Containerized ML workloads with Docker and stored images in Azure Container Registry for consistent deployment across environments. Deployed and orchestrated models on AKS using Helm and Terraform, integrating Prometheus, Grafana, and Azure Monitor for end-to-end observability. Environment: Python, PyTorch, TensorFlow, XGBoost, scikit-learn, FastAPI, Docker, Kubernetes/AKS, Helm, Terraform, Airflow, MLflow, DVC, Azure ML, Azure Functions, Azure Blob Storage, Azure Container Registry, Azure Key Vault, Azure Monitor, Azure DevOps, TensorFlow Serving, TorchServe, Prometheus, Grafana, GitHub Actions Client: Global Atlantic financial group, New York, NY Role: Data Scientist Aug 2018 Jan 2020 Responsibilities: Developed customer churn and credit risk prediction solutions using XGBoost, LightGBM, and scikit-learn, enabling data-driven retention and risk strategies. Designed analytics workflows combining batch scoring, model explanations, and BI reporting to deliver actionable insights to stakeholders. Worked with product, risk, and marketing teams to refine features, thresholds, and rules based on model outputs. Built feature pipelines using Pandas, NumPy, and SQL to process customer demographics, transaction histories, and behavioral signals. Integrated data from MySQL, PostgreSQL, and flat-file sources with robust cleaning, imputation, and normalization steps. Managed curated datasets in relational stores optimized for model training, analytics, and reporting. Implemented data quality checks and validations to ensure stable model inputs and trustworthy insights. Selected algorithms such as XGBoost, LightGBM, logistic regression, and tree ensembles for binary classification and risk scoring. Used feature importance analysis, class-imbalance handling, and probability calibration to align outputs with real-world risk behavior. Ran grid search/random search with cross-validation to maximize AUC, recall, and business KPIs. Deployed models via Flask-based APIs, containerized with Docker, and integrated them into internal decision-support tools. Built EDA reports and visualization dashboards using Seaborn and Plotly, explaining key churn and risk drivers to non-technical audiences. Environment: Python, scikit-learn, XGBoost, LightGBM, Pandas, NumPy, Flask, Docker, spaCy, NLTK, Seaborn, Plotly, MySQL, PostgreSQL, Git, GitHub Client: Trigent Software Bangalore, India Role: Data Engineer Mar 2016 Jun 2018 Responsibilities: Engineered data and analytics automation solutions using Python, Pandas, NumPy, SQL, and Tableau to streamline BI workflows and predictive reporting. Designed ETL architectures to feed downstream analytics and ML models supporting sales forecasting, segmentation, and performance tracking. Built ingestion jobs extracting data from relational databases and CSV/flat files into centralized analytical schemas. Cleaned, transformed, and aggregated raw data using Pandas/NumPy, standardizing metrics and dimensions for dashboards. Managed storage in SQL-based warehouses and optimized query performance using indexing, partitioning, and efficient schema design. Implemented data validation, anomaly detection scripts, and reconciliation checks to ensure high-quality datasets for reporting and analytics. Developed baseline predictive models using scikit-learn (regression, classification, clustering) to support revenue forecasting and customer segmentation. Automated recurring ETL and reporting workflows using Python scripts and schedulers, reducing manual report generation. Built Tableau dashboards for KPI and trend analysis, integrating ML outputs to enhance decision support. Standardized development practices with Git-based version control and modular Python code for maintainability and reuse. Environment: Python, Pandas, NumPy, scikit-learn, SQL, Tableau, Git Eduction : Sathyabama University in Computer science 2015 Keywords: continuous integration continuous deployment artificial intelligence machine learning business intelligence sthree California Connecticut Kansas New York Washington