Home

Satvik Jonnalagadda - Senior AI&ML Engineer

[email protected]

Location: Remote, Remote, USA

Relocation: Yes

Visa: GC Holder

PROFESSIONAL SUMMARY Senior AI/ML Engineer with 11+ years of IT experience, including 8+ years in data engineering and 5+ years shipping production machine learning systems. The most recent focus is on Generative AI, RAG, and agentic systems in HIPAA-regulated healthcare at AbbVie. Career started in traditional data engineering with SSIS, Informatica, and warehouse modeling, then grew with the discipline as it evolved: cloud lakehouses on AWS, Azure, and GCP, large-scale Spark and Kafka pipelines, classical ML in production, deep learning in computer vision and NLP, and now LLM-driven agentic systems. Deep expertise in Python, PySpark, FastAPI, LangChain, LangGraph, PyTorch, and TensorFlow, with end-to-end MLOps and LLMOps lifecycle via MLflow, Weights & Biases, SageMaker, Vertex AI, Docker, Kubernetes, Terraform, and CI/CD. Current AbbVie work covers multimodal RAG with GPT-4o and Claude vision, LangGraph agent orchestration with MCP and native tool/function calling, AWS Bedrock managed fine-tuning on clinical corpora, GPU-accelerated medical-imaging inference, LLM evaluation with LangSmith and Ragas, and Responsible-AI guardrails for clinical documents. Experienced mentoring engineers, leading technical design sessions, and driving AI/ML architecture in fast-paced, cross-functional, matrixed environments. TECHNICAL SKILLS GenAI & Agentic Systems: LLMs, LangChain, LangGraph, Hugging Face, RAG, Vector DBs, Agent Orchestration, Tool/Function Calling, MCP, Prompt Engineering, In-Context Learning, LLM Evals (LangSmith, Ragas), Guardrails / Responsible AI, Managed Foundation Model Fine-tuning (AWS Bedrock) ML & Deep Learning: PyTorch, TensorFlow, Scikit-learn, XGBoost, LightGBM, BERT/Transformers, spaCy, NLP, Computer Vision, CNN/LSTM, Recommender Systems, Time-Series Forecasting (ARIMA, Prophet), Feature Engineering, Optuna, SHAP MLOps & LLMOps: MLflow, Weights & Biases, SageMaker, Vertex AI, Azure ML, Model Registry, Experiment Tracking, Prompt Versioning, Eval Pipelines, Model Monitoring & Drift Detection, NVIDIA Triton, CUDA / GPU Optimization, TensorRT Languages: Python, SQL (T-SQL, PL/SQL), PySpark, Spark SQL, Bash Data Engineering & Streaming: Apache Spark, Delta Lake, Databricks, dbt, Airflow, Apache Kafka, Spark Structured Streaming, AWS Kinesis, GCP Pub/Sub, Hadoop, Hive, Sqoop Cloud Platforms: AWS: S3, Glue, EMR, SageMaker, Lambda, EKS, Bedrock, Redshift, Kinesis Azure: ADF, ADLS Gen2, Synapse, Databricks, Azure ML, DevOps GCP: BigQuery, Vertex AI, Dataflow, Pub/Sub, GKE, Cloud Run Databases & Storage: PostgreSQL, SQL Server, Snowflake, MongoDB, DynamoDB, Cassandra, Redis, ElasticSearch API, CI/CD & Infrastructure: FastAPI, Flask, REST APIs, Docker, Kubernetes (EKS, AKS, GKE), Terraform, Git, GitHub Actions, Jenkins, Azure DevOps, Helm BI & Testing: Tableau, Power BI, Plotly, Matplotlib, Pytest, Great Expectations, LLM Eval Frameworks PROFESSIONAL EXPERIENCE Senior AI/ML EngineerFeb 2025 Present AbbVie | Vernon Hills, IL Designed and shipped a HIPAA-aware, agentic RAG platform over 500K+ biomedical documents (clinical trial protocols, FDA submissions, internal study reports) using LangChain, LangGraph, OpenAI GPT-4o, and Anthropic Claude with hybrid retrieval, reranking, and citation-grounded responses across Pinecone, FAISS, and pgvector indexes, cutting manual literature review effort by ~60% for clinical research and pharmacovigilance teams. Built multi-agent orchestration with LangGraph and Model Context Protocol (MCP) routing compound clinical questions through specialized retrieval, summarization, structured-data lookup, and target-discovery tools via tool/function calling, enabling researchers to ask multi-step questions across unstructured EHR notes, curated trial data, and biomedical knowledge graphs. Implemented multimodal RAG pipelines combining radiology reports, pathology images, and structured EHR fields through vision-language LLMs (GPT-4o, Claude) with Hugging Face embedding models, supporting drug-discovery and clinical research workflows where text alone misses signal buried in scans and figures. Stood up a production LLM evaluation harness using LangSmith, Ragas, and custom eval frameworks that measure retrieval precision/recall, citation faithfulness, hallucination rate, and clinical safety regressions, gating every prompt and model upgrade through versioned eval suites with prompt versioning before promotion. Engineered LLM guardrails and Responsible-AI controls for HIPAA-regulated workloads, including PHI detection and de-identification, prompt-injection filters, output safety classifiers, and human-in-the-loop review gates, keeping every GenAI surface in clinical apps audit-ready. Customized foundation models via AWS Bedrock managed fine-tuning on AbbVie's curated biomedical and adverse-event corpora, lifting downstream extraction F1 by ~15% on AbbVie-specific entities vs. off-the-shelf baselines without managing GPU training infrastructure. Trained and shipped production NLP models on PyTorch, TensorFlow, and BERT/Transformers, including Transformer-based NER and multi-label document classification on EHR data, with spaCy preprocessing pipelines integrated into enterprise document workflows processing 20K+ records weekly. Deployed medical-imaging deep-learning models (CNN, Vision Transformers) for clinical image triage on NVIDIA Triton with CUDA-accelerated GPU optimization on EKS, serving scalable batch and real-time inference across regulated clinical imagery. Engineered FHIR-based clinical data lakes with PHI de-identification logic in PySpark on AWS S3, Glue, and Delta Lake, enabling safe use of clinical data for ML training, fine-tuning, and RAG indexes. Stood up real-time clinical event monitoring pipelines using Kafka, AWS Kinesis, and Spark Structured Streaming over EHR and genomic event streams, delivering sub-minute latency with schema-registry enforcement, dead-letter queue handling, and idempotent processing into Delta Lake and Snowflake orchestrated by Apache Airflow and dbt. Built the full MLOps/LLMOps lifecycle on SageMaker, MLflow, Weights & Biases, EKS, Docker, Helm, Terraform, and GitHub Actions, covering experiment tracking, prompt versioning, model monitoring, drift detection, eval pipelines, staged deployment, and reliable rollback, which shrank release cycles from days to hours and maintained 99.5%+ pipeline uptime. Designed a Patient 360 platform fusing EHR, pharmacy, claims, and trial data on PostgreSQL, DynamoDB, Redis, and ElasticSearch, powering AI-assisted care and research recommendations exposed via FastAPI microservices with Great Expectations and Pytest data-quality and regression coverage. Wired ML and GenAI predictions into three internal clinical applications via FastAPI; published Tableau and Power BI dashboards (with Plotly for ad-hoc analyses) tracking pipeline SLAs, model accuracy, RAG retrieval quality, and clinical KPIs for product and leadership stakeholders. Mentored junior engineers through code reviews, architectural guidance, and knowledge-sharing sessions; led AI/ML solution design with data engineers, architects, scientists, and clinical stakeholders in an Agile, cross-functional matrixed environment. Environment: Python, PySpark, Apache Spark, Kafka, AWS Kinesis, Airflow, dbt, Delta Lake, Snowflake, AWS (S3, Glue, SageMaker, Lambda, EKS, Bedrock), PyTorch, TensorFlow, BERT, Hugging Face, LangChain, LangGraph, MCP (Model Context Protocol), OpenAI GPT-4o, Anthropic Claude (Multimodal), FAISS, Pinecone, pgvector, AWS Bedrock Managed Fine-tuning, MLflow, Weights & Biases, LangSmith, Ragas, NVIDIA Triton, CUDA, Docker, Kubernetes (EKS), Helm, Terraform, FastAPI, PostgreSQL, DynamoDB, Redis, ElasticSearch, Tableau, Power BI, Plotly, GitHub Actions, Great Expectations, Pytest, HIPAA, FHIR AI/ML EngineerOct 2022 Jan 2025 John Deere | Urbandale, IA Trained and deployed deep-learning computer-vision models (CNN, Vision Transformers) for in-field weed detection, crop-health classification, and equipment-perimeter awareness on millions of frames from connected agricultural machinery, supporting precision herbicide application and reduced chemical use across global fleets. Built equipment failure prediction and driver-behavior classification models with Scikit-learn, TensorFlow, PyTorch, XGBoost, and LightGBM on Databricks; reduced unplanned downtime through proactive service scheduling across connected agricultural fleets. Established real-time streaming feature pipelines using Kafka and Spark Structured Streaming over Avro-encoded sensor data from 10K+ active machines with schema-registry enforcement and idempotent consumer design; maintained an ADLS Gen2 + Delta Lake lakehouse as the central ML feature store. Built Azure ADF and PySpark ingestion pipelines processing 50M+ daily IoT telemetry events in JSON, Avro, and Parquet formats from connected equipment, feeding predictive-maintenance, yield-optimization, and route-planning models. Produced K-Means customer segmentation and collaborative-filtering recommender systems ranking implement and parts upsells for dealer-portal personalization across millions of customer-equipment interactions; commercial teams used outputs to personalize service plans and parts replenishment. Developed time-series forecasting models (Prophet, ARIMA, LSTM, LightGBM) for parts demand and seasonal service planning across regional fleets, supporting supply-chain decisions for global parts distribution. Launched NLP pipelines using spaCy and BERT for sentiment analysis and issue classification on service-technician reports and dealer feedback; surfaced insights via Power BI and Tableau dashboards used by product and aftermarket teams. Optimized deep-learning inference for edge and in-cab deployment using TensorRT and CUDA on NVIDIA GPUs, hitting sub-second latency for autonomous decision loops on agricultural equipment. Automated the full MLOps lifecycle on Azure DevOps, MLflow, and Azure ML, covering experiment tracking, Optuna-driven hyperparameter tuning, SHAP-based explainability, model registry, scheduled retraining, drift detection, and staged deployment, which cut time-to-production by ~50%. Integrated GCP BigQuery, Vertex AI, Dataflow, and Pub/Sub for cross-cloud SQL analytics and managed ML workflows; wired model predictions into internal fleet-management apps via FastAPI microservices on Docker and Kubernetes (AKS and GKE) and Cloud Run, exercising AWS, Azure, and GCP within the same architecture. Environment: Python, PySpark, Spark SQL, Apache Spark, Kafka, Azure (ADF, ADLS Gen2, Synapse, Databricks, DevOps, Azure ML), GCP (BigQuery, Vertex AI, Dataflow, Pub/Sub, GKE, Cloud Run), Delta Lake, TensorFlow, PyTorch, Scikit-learn, XGBoost, LightGBM, spaCy, BERT, CNN/LSTM, Vision Transformers, Optuna, SHAP, MLflow, TensorRT, CUDA, Docker, Kubernetes (AKS, GKE), FastAPI, Power BI, Tableau, Agile/Scrum Data ScientistMay 2020 Sep 2022 Edward Jones | St. Louis, MO Built XGBoost-based client attrition / churn prediction models on 2M+ client records, engineering behavioral features from transaction history, advisor touchpoints, and portfolio activity; SHAP-based explanations surfaced top retention drivers, improving early-warning targeting for advisor outreach by ~30%. Developed K-Means and RFM-based client segmentation producing 8 actionable personas across the 2M+ client base; segmentation powered tailored marketing journeys and advisor-engagement plays, lifting campaign response rates by ~25%. Trained product propensity and next-best-action classifiers (XGBoost, LightGBM, logistic regression) ranking advisor leads by likelihood to adopt specific financial products, deployed as weekly batch-scoring jobs on Azure Databricks to feed advisor-facing tooling that informed real-time client conversations. Produced time-series forecasting models for client asset flows, AUM trajectories, and campaign uptake using ARIMA, Prophet, and gradient-boosted regressors; outputs fed regional leadership dashboards used in quarterly planning. Designed an A/B test analysis framework in PySpark to measure campaign and outreach lift, computing statistical significance, effect sizes, and segment-level cuts; results gated go/no-go decisions on retention and cross-sell campaigns. Stood up the modeling lifecycle on Azure Databricks and MLflow, covering feature pipelines, experiment tracking, Optuna-driven hyperparameter tuning, model registry, and scheduled retraining, which cut model handoff and redeployment friction across the data-science team and reduced time-to-production by ~40%. Built NLP pipelines (Scikit-learn, Pandas, classical text features) for issue classification and topic modeling on advisor call-center notes and service complaints, feeding compliance-oriented monitoring workflows. Migrated legacy Hive batch workloads into optimized PySpark and Spark SQL transformations on Azure Databricks, cutting nightly feature-build runtimes from 6+ hours to under 90 minutes and unblocking faster model iteration. Consolidated data from Snowflake, SQL Server, and Oracle into a unified ADLS Gen2 + Synapse platform, retiring 12+ fragmented data silos and standardizing the data foundation for downstream modeling. Wrapped scoring services in Flask APIs with Pytest coverage for internal model consumers; partnered with marketing, compliance, and advisor-enablement teams to translate business problems into modeling specs and present results to non-technical stakeholders, including risk and compliance reviewers; published self-service Tableau dashboards (with Matplotlib for analyses) tracking model performance, campaign uptake, and segmentation health used by 20+ analysts and senior leaders. Environment: Python, PySpark, Spark SQL, Apache Spark, XGBoost, LightGBM, Scikit-learn, SHAP, Optuna, MLflow, Pandas, NumPy, Matplotlib, Azure Databricks, ADLS Gen2, Synapse, Snowflake, Hadoop, Hive, SQL Server, Oracle, Flask, Pytest, Tableau, Git Data EngineerJan 2018 Apr 2020 Sam's Club (Walmart) | Arizona Architected an AWS-based data lake (S3 + EMR + Spark) processing 100M+ weekly retail transaction records across hundreds of stores, replacing an on-prem Hadoop cluster and cutting infrastructure costs. Wired up AWS Glue and Lambda ingestion pipelines pulling from 15+ data sources (POS, third-party vendors, and loyalty platforms), with Apache Kafka streaming for near-real-time inventory and member analytics. Trained and shipped early Scikit-learn member churn and basket-affinity models with collaborative-filtering recommendation prototypes, feeding targeted retention and cross-sell campaigns. This was the first hands-on production ML work that began my pivot from data engineering into ML. Built XGBoost-based demand-forecasting models for category-level replenishment, reducing stockouts and waste across regional distribution centers. Authored PySpark transformation scripts and Hive/Sqoop batch ETL for data cleansing and structured star-schema data marts on AWS Redshift for self-service BI. Built member-360 attribute pipelines (purchase history, store-visit patterns, and loyalty signals) stored across Cassandra, MongoDB, and ElasticSearch for low-latency personalization lookups in member-facing experiences. Set up Jenkins-based CI/CD for data pipelines and exposed early ML scoring through lightweight Flask services; optimized SQL/Hive queries, cutting analytical runtimes by ~45% on distributed Hadoop clusters; launched Tableau dashboards used daily by merchandising and operations leadership. Environment: Python, PySpark, Apache Spark, Hadoop, Hive, AWS (S3, EMR, Glue, Lambda, Redshift), Kafka, Sqoop, Scikit-learn, XGBoost, Pandas, NumPy, Cassandra, MongoDB, ElasticSearch, Flask, Jenkins, Tableau, Git Data EngineerJul 2014 Aug 2017 Careator Technologies Pvt. Ltd. | Hyderabad, India Owned enterprise ETL workflows in SSIS and Informatica PowerCenter, pulling data from SQL Server, flat files, and XML sources across 5+ business units. Built OLAP dimensional models and tuned T-SQL and PL/SQL queries for high-volume operational reporting; automated SSRS report delivery to 50+ business users, cutting manual effort by ~70%. Established data warehouse architecture and ETL documentation standards that reduced pipeline incident rate and accelerated team onboarding; this foundation work set up my later move into cloud and ML. Environment: SQL Server, T-SQL, PL/SQL, SSIS, SSRS, Informatica PowerCenter, Oracle, OLAP, Tableau, Excel, Data Warehousing EDUCATION Bachelor of Technology in Computer Science and Engineering | Indian Institute of Technology, Madras | May 2014 Keywords: continuous integration continuous deployment artificial intelligence machine learning business intelligence sthree active directory information technology golang procedural language Delaware Illinois Iowa Missouri