- Cut diagnostic inference latency 13×, from 2s to 150ms, across 10K+ daily X-ray scans by optimizing PyTorch models with TensorRT and rebuilding the serving layer as FastAPI microservices on edge devices backed by AWS SageMaker — holding 92% precision.
- Raised lung-disease detection accuracy 28% by fine-tuning ResNet and YOLOv8 on labeled clinical imaging datasets and running automated hyperparameter sweeps in Weights & Biases.
- Shortened model release cycles from 4 weeks to 1 week by building end-to-end CI/CD across MLflow, Jenkins, GitHub Actions, Docker and Kubernetes, making every deployment versioned and reproducible.
- Improved cross-site model generalization 15% with zero patient data leaving origin sites, by architecting a HIPAA-compliant federated learning pipeline in Flower spanning 5 regional hospital networks.
- Reduced radiologist draft-review time 35% with an agentic GenAI pipeline — LangChain, LangGraph and RAG over BERT-based LLMs — that generates clinical report drafts directly from imaging model outputs.
- Caught hallucination and drift incidents before release by instrumenting a real-time LLM observability layer with LangSmith, Prometheus and Grafana, paired with bias-detection pipelines and HIPAA fairness dashboards documenting model equity across patient subgroups.
- Processed 500GB+ of imaging data per training cycle by orchestrating Apache Spark and Kafka ETL pipelines with schema enforcement, automated quality validation and augmentation.
- Increased team delivery velocity 20% by mentoring 4 engineers on PyTorch model development, MLOps and LLMOps deployment practice.
AI/ML Engineer · Irving, TX · Authorized to work in the U.S.
Siva PrasadMandalapu
Production ML / MLOps / Generative AI — Healthcare & Financial Services
I build machine learning for systems that get audited. Five years of models in production — HIPAA-regulated clinical imaging at Philips Healthcare, RBI-audited fraud detection at HDFC Bank — plus the pipelines that ship them and the monitoring that keeps them honest.
Inference latency · chest X-ray pipeline
Philips Healthcare · 10K+ scans / day
Current p50
2000ms
Down from 2s. 13× faster, holding 92% precision — PyTorch models compiled with TensorRT, served as FastAPI microservices on edge devices backed by AWS SageMaker.
0×
faster diagnostic inference, 2s to 150ms
Philips
0%
fraud recall across 5M+ daily transactions
HDFC Bank
0%
serving SLA held through peak banking hours
HDFC Bank
0GB
imaging data processed per training cycle
Philips
Production record • 2019 — present
what shipped, and what it moved
Two regulated environments, one job description: get the model out of the notebook and keep it trustworthy once it's there.
- Detected fraud at 95% recall across 5M+ daily transactions with sub-100ms scoring, by deploying Transformer-based NLP models on Azure ML — cutting false positives 30%.
- Reduced fraud losses 35% by engineering LSTM, Transformer and Graph Neural Network ensembles for behavioral anomaly detection over Kafka streaming pipelines.
- Cut fraud-investigation time 40% by implementing SHAP and LIME explainability layers that auto-generated RBI-compliant audit reports for risk and compliance teams.
- Held 99.9% SLA uptime through peak banking hours by automating model retraining, drift detection and rollback with MLflow, Kubeflow and Kubernetes.
- Improved model F1-score 22% by automating feature-engineering ETL across 1M+ labeled transaction records using Pandas and Dask.
- Expanded training data 5× without exposing customer records, by generating synthetic transaction data under differential-privacy constraints and validating it against custom fidelity metrics.
- Delivered 3 fraud-AI systems end to end, leading Agile execution across risk, compliance and engineering from data ingestion through production deployment.
Toolchain
what I reach for
Grouped by the job it does, not by how impressive the logo looks.
Languages5
ML & Deep Learning12
Generative AI & Agents10
MLOps & LLMOps13
Cloud & Serving12
Data Engineering9
Databases & Vector Stores8
Responsible AI & Compliance7
Credentials
Education
-
M.S., Information Systems
Marist College • Poughkeepsie, NY • 2023–2024
-
B.Tech, Mechanical Engineering
Sir C.R.R. College of Engineering • India
Certifications
-
AWS Certified Machine Learning — Specialty
Amazon Web Services
-
TensorFlow Developer Certificate
Google
-
Generative AI with Large Language Models
DeepLearning.AI & AWS
Contact