At Unessa Foundation, I’m developing and deploying AI/ML models for research and technology initiatives. I work on end-to-end machine learning pipelines, from data preprocessing and model training through inference deployment.
I built AgriPredict, a full-stack agricultural decision-support platform with a Flask REST backend and React/Vite frontend serving five specialized ML models. The platform includes crop yield prediction, irrigation scheduling, market price forecasting, sustainability scoring, and crop recommendation.
I designed and trained a 2.3-billion-parameter GPT-style autoregressive language model from scratch using PyTorch. I implemented multi-head self-attention, positional encoding, and transformer decoder blocks.
My projects also include an emotion detection system using a hybrid LSTM+BERT architecture, a fake news classifier, and computer vision work with MediaPipe, OpenCV, and YOLOv8. I’ve also engineered a speaker verification pipeline using TDNN/ECAPA-TDNN, PLDA, and MFCC features.
