Projects

Full-Stack ML App

Recall — Socratic AI Learning Tutor

A mobile learning application that uses an LLM to guide users through book comprehension using the Socratic method — asking questions rather than giving answers.

Socratic prompting system for coherent multi-turn teaching conversations
Book-specific grounding to prevent hallucination and keep responses contextual
Kotlin Spring Boot REST API backend deployed to GCP Cloud Run
Supabase (PostgreSQL + Auth) + Groq API for fast inference
Compose Multiplatform client (Android + iOS)
Chrome extension for book content extraction

Kotlin Spring Boot Jetpack Compose Compose Multiplatform GCP Cloud Run Supabase PostgreSQL Groq API LLM

In progress — targeting release as a fully end-to-end shipped product

ML Projects

Bean Disease Classification (Computer Vision)

Full end-to-end pipeline from training to mobile deployment.

Deep CNN with transfer learning, 98% test accuracy on ~1k samples
Data augmentation to compensate for small dataset
MLflow experiment tracking + Apache Airflow workflow orchestration
TFLite deployment to Android for real-time offline inference
Dockerized Modal cloud GPU training for scalable experimentation
Flask REST API for configurable model training and hyperparameter control

Keras TensorFlow TFDS TFLite MLflow Airflow Docker Flask HuggingFace Android

GitHub | Hugging Face

Air Quality Prediction — Time-Series Forecasting

Comparative forecasting study with a full MLOps setup.

Compared ARIMA, Prophet, XGBoost, and LSTM models (best Willmott Index: 0.84)
Recursive multi-step forecasting with rolling-window cross-validation
CI/CD pipeline with GitHub Actions and Docker

Python Scikit-Learn XGBoost TensorFlow/LSTM Prophet statsmodels Docker GitHub Actions

GitHub

Sudoku Solver with Deep CNNs

TFRecords for efficient large-dataset loading and preprocessing
Curriculum (progressive) learning — trained on easier puzzles first, gradually increasing difficulty
Custom loss function and adapted architecture to enforce Sudoku rules

Keras TensorFlow TFRecords Deep CNNs