- Cut inference latency 13× (2s to 150ms per scan) at 10K+ scans a day with no loss in precision, by converting PyTorch models to TensorRT and rebuilding model serving as containerized FastAPI services on on-prem GPU edge devices.
- Cut model release-candidate turnaround from 4 weeks to 1 by building ML CI/CD with GitHub Actions, Jenkins, Docker and Kubernetes, an MLflow model registry, and regression tests on a locked validation set.
- Monitored drift and model health for the 10K+ scans/day service with Prometheus and Grafana, tracking data drift, service health and per-subgroup model performance to give early visibility into model and service issues.
- Made every training run reproducible over 500GB+ of imaging data with Spark and Kafka pipelines, schema validation and automated data-quality gates feeding AWS SageMaker training jobs.
- Trained models across 5 hospitals with zero patient images moved, by building the client-side training and evaluation pipeline for a federated learning pilot (Flower).
- Helped cut radiologist review time 35% in an LLM pilot by building the LLMOps layer for a LangGraph report-drafting pipeline on Azure OpenAI: LangSmith tracing and automated checks for unsupported statements.
Machine Learning Engineer · Irving, TX · Authorized to work in the U.S.
Siva PrasadMandalapu
Production ML, MLOps, Fraud and Healthcare AI
I get models into production and keep them healthy there. Five years running real-time ML in regulated industries: 5M+ fraud scores a day at 99.9% uptime at HDFC Bank, and 13× faster medical-imaging inference at Philips Healthcare. AWS Certified Machine Learning – Specialty.
Per-scan latency
2000ms
Down from 2s. 13× faster at 10K+ scans a day, no loss in precision. PyTorch models converted to TensorRT, served as containerized FastAPI on on-prem GPU edge devices.
Scored since you got here
0
At 5M+ a day that's ~58 every second, each under 100 ms at 99.9% uptime, on containerized endpoints on Azure ML and Kubernetes.
Counter runs at the real daily rate
Before
After
commit→
GitHub Actions→
Jenkins→
Docker build→
regression tests · locked validation set→
MLflow registry→
Kubernetes
Release-candidate turnaround
1week
Down from 4 weeks. ML CI/CD with GitHub Actions, Jenkins, Docker and Kubernetes, an MLflow model registry, and regression tests on a locked validation set.
0×
faster medical-imaging inference, 2s to 150ms
Philips
0M+
real-time fraud scores a day, under 100 ms each
HDFC Bank
0%
uptime on fraud-scoring endpoints
HDFC Bank
0%
lower fraud losses, with 30% fewer false positives
HDFC Bank
The loop • click a stage
every stage of a model's life, shipped twice
Data to monitoring and back again, once for medical imaging and once for fraud. Pick a stage to see what I built there.
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. Filter by company, tool or keyword.
- Served 5M+ real-time fraud scores a day at under 100 ms and 99.9% uptime by deploying containerized model endpoints on Azure ML and Kubernetes.
- Turned model updates into scheduled, automated releases by building Kubeflow retraining pipelines with MLflow experiment tracking and model versioning, feature and score drift detection, and automated rollback.
- Cut false-positive alerts 30% at 95% recall and fraud losses 35% by partnering with data scientists and fraud analysts to productionize an LSTM, Transformer and graph neural network ensemble.
- Cut fraud investigation time 40% by attaching SHAP explanations to every flagged transaction and auto-generating case summaries in the format required by RBI auditors.
- Improved model F1 22% with Pandas and Dask feature engineering pipelines over 1M+ labeled transactions.
Nothing in the record matches that.
Toolchain
what I reach for
Grouped by the job it does. Click any tool with a count to see where it shows up in the record.
MLOps
Model Serving
Observability
Data & Cloud
ML & Languages
Count n = results in the record that used it.
Credentials
Education
- Jan 2023 – Dec 2024
M.S., Information Systems
Marist College • Poughkeepsie, NY
- Jun 2015 – May 2019
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