CoralSense AI
Ongoing · Oceanic AI Lab · Team of two
Problem: Reef-health surveys are expensive and sparse, so bleaching often goes unrecorded.
Approach: A low-cost drifting buoy with underwater imaging, temperature/pH/salinity sensing and GPS. I'm building the AI pipeline: underwater image enhancement followed by U-Net segmentation of coral as Healthy, Bleached or Dead, trained on Coralscapes (2,075 annotated images).
Result: Geo-referenced reef-health records; presented orally at OCEANAI-2026. Software stage nearing completion, hardware integration next.
Image coming soon
Spoken Language Identification
NIT Goa Research Internship · Summer 2026
Problem: Identify which of 8 Indian languages is spoken in an audio recording.
Approach: MFCC features from the IIT Mandi ReadSpeech dataset; 8 classifiers benchmarked with 5-fold cross-validation and confusion-matrix analysis.
Result: Tuned SVM reached 91.99% test accuracy, deployed in a Flask web app.
Image coming soon
CVV Hostel E-Outpass System
Completed 2026
Problem: Hostel outpasses ran on paper, email and WhatsApp, with no reliable approval trail.
Approach: Parent OTP login, role-based approval routing to teachers and wardens by academic-calendar rules, and QR e-outpasses scanned at the main gate for exit and return.
Result: Submitted to the college administration for adoption as the official outpass system.
Image coming soon
Crop Yield Prediction
Problem: Predict crop yields, and say how confident each prediction is.
Approach: End-to-end ML pipeline on 1M+ rows with Linear Regression and XGBRFRegressor, extended with conformal prediction intervals.
Result: R² ≈ 0.91, deployed as a Flask web app; the uncertainty work became a BDA 2026 paper.
