AI/ML Engineer: LLM, RAG, Agentic Systems

SANMITHA NG

234-***-**** ************@*****.*** LinkedIn Seattle, WA Open to Relocation

PROFESSIONAL SUMMARY

AI/ML Engineer with 4+ years delivering production LLM systems, RAG pipelines, and agentic AI workflows across financial

services and healthcare. Core stack: PyTorch, Hugging Face Transformers, LangGraph, FinBERT, BioBERT, AWS SageMaker,

Azure ML, MLflow, and vLLM. Proven impact: 22% NLP precision lift, 40% latency reduction, SR 11-7 and HIPAA compliant

deployments at enterprise scale across JP Morgan Chase and Accenture.

AREAS OF EXPERTISE

LLM Fine-Tuning & Instruction Tuning Retrieval-Augmented Generation (RAG) Agentic AI Workflows (LangGraph)

Credit Risk & AML Modeling Clinical NLP & EHR Integration Prompt Engineering & Evals (Ragas, LangSmith) MLOps

& LLMOps Model Risk Governance (SR 11-7) HIPAA & PHI Compliance Inference Latency Optimization (vLLM)

PROFESSIONAL EXPERIENCE

AI/ML Engineer 01/2025 – Present

JP Morgan Chase, USA

• Designed stateful multi-agent LLM workflows using LangGraph with tool-calling across 1M+ daily trade and credit records,

reducing regulatory report turnaround 20% for 200+ risk and compliance stakeholders.

• Fine-tuned LLaMA-3 and FinBERT on proprietary financial corpora via Hugging Face with QLoRA instruction tuning,

lifting credit risk NLP precision 22% across loan covenant extraction and AML transaction classification.

• Architected RAG pipelines over SEC filings and trade surveillance logs using Pinecone and FAISS with prompt guardrails

and context reranking, cutting hallucination rate from 6.1% to 0.8% for 200+ analysts.

• Deployed end-to-end ML pipelines on AWS SageMaker and Azure ML with vLLM inference serving, reducing pipeline cycle

times 40%; implemented MLflow drift detection covering SR 11-7 model risk governance across 3 business units.

• Instrumented LLM observability via LangSmith tracing and dbt test suites covering 95% of financial data transformations,

reducing production data quality incidents 32% and enabling explainable AI audit trails for compliance.

• Built XGBoost and LightGBM credit default models on Snowflake loan portfolio data, surfacing predictions into Power BI

KRI dashboards and improving early warning signal accuracy 18% over the prior logistic regression baseline.

ML Engineer 02/2021 – 07/2024

Accenture, India

• Trained XGBoost and Random Forest patient readmission risk models on 5M+ EHR records ingested via FHIR and HL7

APIs, improving early intervention F1-score 13% over the prior rule-based baseline across 4 hospital units.

• Developed PyTorch LSTM and CNN architectures for clinical time-series modeling of patient vitals and lab sequences,

applying transfer learning and cross-validation to reduce model training time 28% across 20+ deployment pipelines.

• Built NLP pipelines using BioBERT and ClinicalBERT for medical NER, ICD-10 extraction, and discharge summary

classification; deployed via FastAPI microservices processing 10K+ daily clinical notes with 85%+ precision and recall.

• Containerized clinical ML services with Docker and Kubernetes on AWS; implemented MLflow experiment tracking and

Isolation Forest anomaly detection on claims data, improving billing fraud detection precision 25% and reducing deployment

errors 15%.

• Applied patient cohort segmentation and A/B testing to validate model improvements before rollout; built SQL population

health dashboards in Tableau improving outcomes reporting accuracy 27% for 3 health system clients.

Data Analyst 05/2020 – 01/2021

Genpact, India

• Built Python and SQL analytics pipelines across 4 source systems, delivering KPI dashboards and data quality reports that

reduced manual reporting effort 35% for 10+ business stakeholders and surfaced downstream ML feature signals.

• Conducted exploratory data analysis and feature engineering using Pandas and NumPy on financial datasets, identifying

predictive signals that improved subsequent ML model accuracy and pipeline integrity across production transformation

workflows.

• Automated recurring financial reconciliation reports using Python scripts and Excel macros, eliminating 6 hours of weekly

manual processing and improving data accuracy 30% across accounts payable and receivable teams.

• Collaborated with business analysts and operations leads to translate reporting requirements into structured SQL queries

and Tableau dashboards, improving visibility into cost variance trends for 5 cross-functional finance teams.

PROJECTS

Agentic Fraud Detection AI System Python, XGBoost, LangGraph, LLM, Isolation Forest, Pinecone, Tableau,

Ragas

• Architected a multi-agent fraud detection pipeline on 5M+ transaction records combining XGBoost, Isolation Forest, and

LLM-generated risk summaries via LangGraph agents with tool-calling, achieving 96% precision and strong F1-score.

• Ran Ragas-based hallucination evals on LLM risk narratives, reducing hallucination rate to under 1% and establishing a

CI-integrated prompt regression test suite for automated quality gates before every production deployment.

• Deployed real-time Tableau KPI dashboard with Plotly visualizations and automated alert routing, reducing manual fraud

review overhead 40% while maintaining model governance and responsible AI audit documentation for compliance teams.

RAG Clinical Decision Support System Python, RAG, FAISS, Hugging Face, LangChain, LangSmith,

XGBoost, FastAPI, HIPAA

• Built a production RAG clinical Q&A assistant using FAISS vector store and Hugging Face LLMs with LangChain

orchestration, enabling context-aware decision support from structured patient records with optimized inference latency.

• Integrated LangSmith tracing for hallucination monitoring and groundedness scoring, reducing hallucination rate from 5.8%

to 0.9% and establishing nightly Ragas-based eval coverage across 3 clinical knowledge repositories.

• Combined XGBoost and Random Forest HRV biomarker models achieving 94% diagnostic accuracy, a 12%+ lift over

baseline, while enforcing HIPAA and PHI compliance via role-based access controls on all patient data pipelines.

EDUCATION

Master of Science in Data Science 08/2024 – 05/2026

Kent State University, Kent, OH

Bachelor of Engineering in Information Science 07/2019 – 06/2023

Jawaharlal Nehru National College of Engineering, India

CERTIFICATIONS

• IBM Data Analysis with Python

• Google Data Analytics

• Microsoft AI Dashboards

• AWS Cloud Practitioner Essentials

TECHNICAL SKILLS

Programming & Frameworks: Python, SQL, R, Bash, PyTorch, TensorFlow, Scikit-learn, LangChain, Hugging Face

Transformers, FastAPI, Flask

Generative AI & LLMs: LLM Fine-Tuning (QLoRA, LoRA), FinBERT, BioBERT, ClinicalBERT, LLaMA-3, RAG

Pipelines, Prompt Engineering, Agentic Workflows (LangGraph, CrewAI), Tool Calling, Model Eval (Ragas, LangSmith),

Guardrails

ML & Deep Learning: XGBoost, LightGBM, Random Forest, CNNs, RNNs, LSTMs, Transformers, Transfer Learning,

NLP (Medical NER, ICD-10 Coding, AML Classification, Sentiment Analysis), Anomaly Detection, Time Series Forecasting,

Feature Engineering, A/B Testing

LLMOps & MLOps: Azure ML, AWS (S3, EC2, Lambda, SageMaker), MLflow, Airflow, Docker, Kubernetes, vLLM,

CI/CD, Drift Detection, Vector Stores (Pinecone, FAISS), Embedding Generation, GPU Acceleration (NVIDIA A100/H100)

Data Systems: Snowflake, AWS Redshift, Azure Databricks, PostgreSQL, MongoDB, Redis, BigQuery, dbt, Spark, Informatica

Visualization & BI: Tableau, Power BI (DAX, KRI Dashboards), Matplotlib, Seaborn, Plotly

Domain & Compliance: Credit Risk Modeling, AML, Trade Surveillance, Model Risk Governance (SR 11-7), FHIR, HL7,

EHR Integration, HIPAA & PHI Compliance, ICD-10, Responsible AI

Collaboration & Agile: Agile/Scrum, Sprint Planning, Stakeholder Communication, Cross-Functional Team Collaboration,

JIRA, Confluence, Git, GitHub Copilot, Code Review, SDLC

Contact this candidate