Bhargav Sairam

AI/ML Intern · Colorado Springs, CO

• Lead development of an end-to-end geospatial ML pipeline fusing five open-source signals into a composite risk score, narrowing a 41,600-cell search grid to 160 priority cells for satellite tasking.
• Engineered a 4-channel ResNet-50 classifier (RGB + near-infrared Sentinel-2 imagery) with transfer learning and class-imbalance loss weighting, validated with 5-fold spatial cross-validation to eliminate geographic leakage.
• Designed a geometry-based cascade filter (NDVI/NDWI spectral masking + shape descriptors) that cut false positives 95% downstream of the CNN.

Software Engineering Intern, AI/ERP Integration · Denver, CO

• Built and deployed a data integration pipeline linking NetSuite ERP with Google Cloud SQL via RESTlet APIs and OAuth 1.0.
• Automated daily transactional sync with Python, MySQL and Cloud Scheduler, cutting manual reporting effort by 90%.
• Designed and optimized staging databases for faster ERP analysis and downstream AI/BI applications.

Machine Learning & Research Intern · Golden, CO

• Researched and implemented ML models to predict resident behaviors and health risks in assisted-living facilities.
• Prototyped an early-warning system that flags potential incidents (falls, deteriorating health) so caregivers can intervene proactively.
• Authored and published a research paper translating ML methods into practical healthcare applications, validated with caregivers and staff.

Land-Cover Classification from Sentinel-2PyTorch, ONNX Runtime Web, Sentinel-2, Web Workers

A ResNet-18 fine-tuned on satellite imagery, quantized to int8 and running entirely in your browser, with class-activation heatmaps that show where the model looks.

MatchFlixNode.js, Express, PostgreSQL, Docker

A swipe-based movie discovery app where groups match on what to watch together, with LLM-powered recommendations from swipe behavior.

Facial Recognition AttendancePython, OpenCV, face-recognition, NumPy

Replaced manual sign-in at KTP meetings with face detection and identity verification, making attendance up to 80% faster and more accurate.