Home

Devika - AI ML ENGINEER

[email protected]

Location: Sunnyvale, California, USA

Relocation: yes

Visa: GC

Devika [email protected] |+1 (408)-459-9307 GenAI | AI/ML Engineer | NLP | AWS PROFESSIONAL SUMMARY GenAI and AI/ML Engineer with 10+ years of experience designing and deploying scalable machine learning, data engineering, and GenAI solutions across AWS and Azure environments. Strong expertise in Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), prompt engineering, and agentic AI systems using frameworks like Lang Chain and Lang Graph. Experienced in improving model performance through supervised fine-tuning (SFT), structured data curation, and advanced prompt optimization techniques, with a focus on building reliable and context-aware AI applications. I am skilled in developing end-to-end ML pipelines, including data ingestion, feature engineering, model training, deployment, and monitoring using tools such as PySpark, MLflow, Docker, and Kubernetes. Proficient in designing scalable ETL/ELT pipelines, optimizing large-scale SQL queries, and building batch and real-time data processing systems using Databricks, Kafka, and cloud-native services. Strong background in backend development using Python (Fast API, Flask) and implementing robust CI/CD and MLOps workflows. Developed AI-driven customer support automation solutions, including query resolution, ticket summarization, and knowledge retrieval for enterprise support workflows. Demonstrated ability to work in fast-paced, cross-functional environments, leading projects from concept to production and delivering high-impact AI-driven solutions that enhance business outcomes. Experience building secure and compliant AI systems with a focus on data privacy and model safety. Strong experience in document AI pipelines, including OCR-based extraction, document classification, and entity recognition from unstructured data. Hands-on experience integrating AWS AI services such as Textract, Comprehend, and Bedrock for scalable document processing solutions Experience defining enterprise AI architecture, orchestration frameworks, and scalable deployment strategies across hybrid cloud (Azure + AWS). Strong exposure to AI governance, model evaluation, compliance considerations, and secure AI system design. Experienced in building agentic AI workflows for automation use cases aligned with enterprise operations and IT processes. CORE SKILLS & TECHNOLOGIES AI/ML Architectures: Traditional Machine Learning & Deep Learning: Extensive experience with ML algorithms including SVM, Random Forest, XGBoost, k-NN, and deep learning architectures such as CNNs, RNNs, LSTMs, and GRUs for classification, regression, clustering, forecasting, and time series prediction. I am proficient in Scikit-learning, XGBoost, LightGBM, Keras, and TensorFlow. Transformers: Hands-on experience implementing and customizing transformer models using Hugging Face, PyTorch, and TensorFlow for tasks such as text classification, question answering, summarization, translation, and multimodal applications. Generative Models: GANs: DCGAN, StyleGAN variants for synthetic image generation, data augmentation, and domain adaptation. VAEs: Variational Autoencoders for anomaly detection and representation learning. Diffusion Models: Experience with Stable Diffusion pipelines and Denoising Diffusion Probabilistic Models (DDPMs) for image and text generation. Large Language Models (LLMs): GPT, Claude (Anthropic via AWS Bedrock), LLaMA, and other open-source LLMs Multi-Agent Systems & Orchestration: Expertise in building agentic workflows using Lang Chain, AutoGen, and Crew AI, including tool integration, agent-to-agent communication, memory management (vector stores, ephemeral and persistent memory), and complex reasoning workflows. NLP Pipelines & RAG: Strong knowledge of NLP pipelines including tokenization (Word Piece, BPE), embeddings (static, contextual, sentence embeddings), attention mechanisms, and Retrieval-Augmented Generation (RAG) using vector databases such as Pinecone, Weaviate, Redis, Milvus, and FAISS. GenAI Evaluation & Validation: Experience designing evaluation frameworks using metrics such as perplexity, BLEU, ROUGE, accuracy, and F1-score, along with human evaluation, adversarial testing, bias detection and mitigation, model calibration, and A/B testing. Conversational AI & Prompt Engineering: Skilled in designing stateful conversational workflows using Lang Graph, advanced prompt engineering, instruction tuning, and context injection for enterprise-grade AI applications. Experience building conversational systems similar to Google CCAI using LLM-based architectures, intent handling, multi-turn dialogue management, and backend integrations. Agentic Workflow Automation: Experience building intelligent automation systems using Lang Chain agents, vector search, and external API orchestration. Azure AI Services: Hands-on experience with Azure Cognitive Services, including Language Studio, Azure AI Search, and Azure Machine Learning for LLMOps. Document AI & NLP Extraction: Built document processing pipelines for OCR-based text extraction, document classification, and entity recognition Developed NLP/LLM-based solutions to convert unstructured documents into structured data outputs Experience working with AWS Textract (OCR) and Comprehend (NER, key phrase extraction) for document understanding Designed workflows for text summarization, metadata extraction, and content classification Cloud & AI Platform Architecture: Azure AI / Microsoft Foundry (AI Studio equivalent), multi-agent systems & AI workflow orchestration, multi-region / multi-environment deployments, Infrastructure-as-Code (Terraform, Ansible exposure) CI/CD & environment promotion (dev/test/prod) ML Engineering & MLOps: Model Lifecycle Management: End-to-end experience in data ingestion, preprocessing, feature engineering (Python, PySpark), experiment tracking (MLflow), hyperparameter tuning, model optimization, deployment (Azure ML, AWS SageMaker, Kubernetes), and monitoring using Prometheus and custom dashboards. Pipeline Orchestration: Experience with Kubeflow Pipelines, Apache Airflow, and Azure Data Factory for orchestrating training and inference workflows. Containerization & CI/CD: Proficient in Docker, Kubernetes, Azure DevOps, and GitHub Actions for continuous integration and deployment of ML and AI services. Backend Development & APIs: Strong experience in Python-based backend development using Fast API, Flask, and Django, including RESTful API design and WebSocket-based real-time communication. Data Engineering: Expertise in SQL (Teradata, MySQL, PostgreSQL), ETL processes using Azure Data Factory and Informatica, data lakes (Azure Data Lake, AWS S3), data warehousing, and real-time data streaming using Kafka, Pulsar, and NATS. Tools & Frameworks: Machine Learning & AI: PyTorch, TensorFlow/Keras, Scikit-learn, Hugging Face Transformers, Diffusers, OpenAI API, LangChain, AutoGen, Crew AI. Computer Vision: OpenCV, YOLO (v5/v8), SAM, DINOv2, ResNet-based embedding pipelines, and NVIDIA DeepStream for real-time object detection and tracking. Programming: Python (Primary), JavaScript (for AI interfaces), basic exposure to C/C++ for performance optimization Vector Databases: Pinecone, Weaviate, Redis Vector Search, FAISS. Cloud Platforms: Azure (Azure ML, Functions, Data Factory, Cognitive Services) and AWS (Lex, Connect, AppSync, API Gateway, Lambda, EC2, S3, CloudWatch, DynamoDB, EKS, RDS, Load Balancing, Auto Scaling, CDN, Networking, SageMaker). Monitoring & Logging: Prometheus, Grafana, Azure Monitor, AWS CloudWatch. AWS Services: AWS Glue, AWS CloudWatch, S3, EC2, Lambda Optimization: AWS Cost Optimization, Performance Tuning, Query Optimization, Batch Processing Optimization Version Control & Collaboration: Git, GitHub, GitLab, Jira, and Agile/Scrum methodologies. EDUCATION Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science and Technology, Chennai, TN GPA: 3.8/4 Bachelor of Technology: Computer Science June 2011 - May 2015 EXPERIENCE Fiserv | Sunnyvale, CA Jan 2023 Present Lead AI/ML Engineer Conversational AI for Store Advising Developed an AI-powered chatbot to assist in-store customers with product recommendations, plan comparisons, and purchasing decisions. Led the development of chatbot functionality, ensuring seamless storage and retrieval of conversation history using database systems. Designed and implemented LangGraph-based conversational workflows to manage dynamic, multi-turn dialogues with branching logic for recommendations, upselling, and comparisons. Experience working in Linux environments and deploying applications using Kubernetes. Performed fine-tuning and experimentation with OpenAI GPT-3 models (Curie, Davinci, and Ada) to enhance conversational accuracy and performance. Analyzed AWS resource utilization and identified cost hotspots, optimizing compute usage and reducing unnecessary processing overhead Designed and implemented Customer Data Platform (CDP)-like solutions for customer identity resolution, profile unification, and data integration across multiple enterprise systems Tuned PySpark data pipelines and SQL queries to improve execution performance and reduce batch processing time Configured and optimized AWS Glue jobs for efficient ETL processing and cost-effective execution Collaborated with cross-functional teams to validate optimizations and ensure measurable improvements in system efficiency Applied supervised fine-tuning (SFT) techniques on LLMs using curated datasets to improve response quality and align model outputs with expected behavior Built agentic AI systems to automate enterprise workflows, reporting, and operational processes Designed solutions aligned with enterprise automation and IT workflow optimization use cases Integrated APIs and backend systems to support end-to-end orchestration and automation pipelines Worked on AI safety and alignment by implementing guardrails, bias mitigation strategies, and controlled response generation to reduce harmful outputs Designed conversational workflows with intent classification, entity extraction, and context management similar to Dialogflow CX architecture. Implemented multi-turn dialogue systems with state management, fallback handling, and escalation strategies. LLM-Driven Chatbot for SQL Generation and Visual Analytics Developed a multi-agent conversational system using Lang Chain and Lang Graph, Integrating Claude (Anthropic via AWS Bedrock), Titan, and GPT-4 foundation models for enterprise-grade AI solutions, and open-source LLMs to convert natural language queries into optimized SQL and analytical insights. Leveraged AWS Bedrock to deploy and scale Claude (Anthropic) and other foundation models for high-performance AI applications Optimized SQL query generation for complex joins and aggregations, ensuring support for enterprise-scale datasets. Developed entity resolution and record linkage solutions using similarity matching techniques and clustering algorithms to unify customer profiles across multiple data sources Applied fuzzy matching techniques (Levenshtein distance, Jaccard similarity) for attribute-level comparison including names, identifiers, and structured data Built embedding-based similarity systems using transformer models and vector search (FAISS) for semantic entity matching Designed scalable pipelines for deduplication and entity matching across large datasets using PySpark and distributed processing Familiar with GCP services such as BigQuery and conversational AI ecosystems; capable of quickly adapting to platforms like Google Dialogflow and CCAI. Intent classification, entity recognition, context handling, session management, and conversation state tracking. State of Louisiana (Methods Technology Solutions) | Baton Rouge, LA Jan 2021 Jan 2023 Lead Data Scientist Insights Module for School Caf Developed an Insights Module for the School Caf platform, delivering predictive analytics, forecasting, and KPI dashboards aggregated from weekly to annual levels for revenue, meals (MEQs), participation, production accuracy, and wastage, contributing to increased platform adoption and sales. Led a team of five to develop and train machine learning models that predict menu item consumption, participation rates, and sitelevel sales using student and operational data. Spearheaded the project end-to-end, including requirements, gathering, use case analysis, data collection, model development, backend engineering, and UI integration for the web platform. Developed KPI visualizations using Plotly and deployed them to Power BI Cloud for user interface integration, including thresholdbased alerting for enhanced usability. Built predictive models to identify students eligible for free and reduced meal programs and forecast their likelihood of participation. Designed and scheduled data pipelines using Azure Data Factory and Databricks (PySpark), enabling hourly data processing and creating a centralized staging layer in the data lake for cross-team access. Wrote highly optimized SQL queries to aggregate millions of records across multiple timeframes, supporting both predictive dashboards and machine learning pipelines. Developed backend applications using the Django framework, containerized with Docker, and deployed through Azure DevOps CI/CD pipelines. Optimized large-scale batch data pipelines using PySpark and Azure Data Factory, improving performance and reducing compute costs Tuned SQL queries and aggregation pipelines for high-performance analytics on large datasets Utilized MLflow for experiment tracking, model management, and streamlined deployment of forecasting models into production dashboards. Applied simulation-based modeling and statistical analysis to improve forecasting accuracy and decision-making Implemented Infrastructure as Code using Terraform to provision and manage Azure resources, ensuring scalability and consistency across environments. Built Grafana dashboards to monitor pipeline latency, job success/failure rates, and resource utilization for proactive system monitoring. Collaborated with stakeholders regularly to understand business requirements, identify pain points, and continuously improve application performance and user experience. Digit7 | Dallas, TX July 2019 Dec 2020 Data Scientist Digit kart Contributed to the development of an automated retail checkout system using multi-camera computer vision models and scalable data pipelines, improving checkout efficiency and customer experience. Developed and optimized computer vision models, including object detection and semantic segmentation, for accurate real-time product identification. Evaluated and integrated multiple AI/ML frameworks and platforms based on performance, scalability, and enterprise requirements Designed and evaluated multiple convolutional neural network (CNN) architectures to identify the best-performing models under hardware and cloud constraints. Integrated communication protocols, including WebSocket s, MQTT, and REST APIs, to enable real-time product listing and interaction on edge device displays. Developed algorithms to estimate product dimensions (e.g., height) and trigger alerts for items are not present in the product database. Collaborated with a cross-functional team to containerize applications, automate edge device deployments, and support hardware design and processing unit optimization. Implemented continuous learning strategies and built data pipelines to enable model adaptability to evolving product data and environments. Designed and implemented end-of-to-end data pipelines on Azure to support seamless onboarding of new clients. Developed SQL-based ETL processes for preprocessing product metadata, transaction logs, and real-time inventory data within Azurebased pipelines. Utilized MLflow for experiment tracking, model versioning, and reproducible deployments across multiple client environments. Automated cloud infrastructure provisioning using Terraform, enabling scalable deployment of computer, storage, and networking resources. 7-Eleven | Dallas, TX Jan 2018 July 2019 Data Scientist Credit Card Fraud Detection Analyzed large-scale transactional datasets (1M+ records) to detect fraudulent credit card transactions using classification models in Python. Addressed class imbalance using advanced resampling techniques, including SMOTE, over sampling, under sampling, and cluster-based sampling methods. Performed extensive data preprocessing, including duplicate removal, outlier detection, missing value imputation (k-NN, linear regression), and correction of mislabeled data. Engineered SQL-based feature pipelines from transactional history, enabling near real-time fraud detection across millions of records. Implemented similarity-based algorithms (cosine similarity and k-NN) to compare new transactions with historical fraud patterns and identify anomalies. Utilized MLflow for experiment tracking and model management, optimizing models based on precision, recall, and F1-score across multiple sampling strategies. Automated infrastructure provisioning using Terraform to dynamically scale compute resources for model training and batch inference. Monitored inference pipelines using Prometheus and Grafana to ensure low latency and high reliability in fraud detection systems. Master Data Management Develop an application for organizations to integrate and aggregate data that help you understand customer better. Also built a web platform to help business users deploy and maintain ML models easily without any use of any code. Our ML solution can be used to cluster similar data to find duplicate records, which provides better data decisions at organizational level. Utilized search algorithms such as approximate nearest neighbor (ANN) and fuzzy string matching to identify semantically similar entries and enhance record linkage accuracy. Integrated MLflow to streamline deployment of clustering models and track updates as organizational datasets evolved. Terraform used for automating cloud-based environments, required for large-scale data aggregation and search workflows. Added Grafana dashboards to monitor API latency, clustering job runtimes, and memory consumption, ensuring platform reliability. Customer Chat Analysis Applied Sentimental Analysis on Customer chat data, Built Word Clouds, Word Correlations, was able to extract Customer Messages and Agent Messages separately from the raw data. Extracted Sentiment and Entities from extracted messages and Built Tableau Dashboards to easily look up entities and chats using different filters to find out the new issues and important feedback from customers. Utilized advanced SQL queries to extract temporal patterns, sentiment shifts, and frequently occurring issues across millions of chat records. Processed the Raw data into organized form by applying Data Cleaning techniques using libraries like pandas and NumPy. Web Scraping and Data Analysis Web scraping various medical related websites, papers related to cure of Cancer using Beautiful Soup in Python. Analyzed the scarped text, performed cluster analysis and visualized the categories in D3.js. Analyzed large dataset of 50M rows of text from various news channels to predict the news veracity quickly. Performed feature selection, Tf-idf, topic modelling, sentimental analysis followed by active learning-based CNN with 2 orders less data with almost 86% accuracy. Used Amazon AWS EC2 and S3 to run the model. Structured scraped datasets using SQL databases for efficient querying, sentiment labeling, and downstream NLP pipeline integration. Star Health Insurance | Chennai, Tamil Nadu Jul 2015 Aug 2017 Python Developer Predictive Analytics & Machine Learning Analyzed large-scale healthcare and workers compensation datasets to predict member risk, treatment outcomes, and cost drivers. Developed and evaluated machine learning models including Random Forest, XGBoost, and Support Vector Machines, improving prediction accuracy and supporting proactive care strategies. Engineered features from claims data, member demographics, and clinical history to enhance model performance and interpretability. Applied cross-validation and performance metrics (ROC-AUC, precision, recall) ensure model robustness and reliability. Translated analytical findings into actionable insights for business stakeholders, enabling data-driven decision-making. Data Cleaning, Transformation & Preparation Performed end-to-end data preprocessing including handling missing values, duplicates, outliers, and inconsistent formats across multiple data sources. Improved data quality and consistency through systematic validation and transformation processes using SQL and Python. Automated data cleaning and transformation workflows, reducing manual effort and improving processing efficiency. Standardized datasets for downstream analytics, reporting, and machine learning pipelines. ETL Optimization & Data Quality (Informatica) Diagnosed and resolved failures in Informatica ETL pipelines caused by post-upgrade data quality issues. Identified root causes include duplicate records, incorrect active indicators, and schema inconsistencies across systems. Developed optimized SQL queries, stored procedures, and macros to cleanse and reconcile data across pipelines. Enhanced ETL performance and reliability by implementing validation checks and error-handling mechanisms. Increased ETL job success rates and reduced downtime by stabilizing data ingestion workflows. Experience working with agentic orchestration platforms and frameworks (Lang Chain, AutoGen, AWS Bedrock Agents) similar enterprise tools like Workato, AtomicWork, and Harmony.ai Worked with healthcare and claims datasets, including structured and unstructured medical records, for predictive modeling and risk analysis Processed document-heavy datasets such as claims and reports to extract meaningful insights Keywords: cprogramm cplusplus continuous integration continuous deployment artificial intelligence machine learning user interface javascript access management business intelligence sthree rlang information technology trade national California Louisiana Tennessee Texas