Vinay Matroju

A cloud retail analytics platform that predicts customer lifetime value, flags churn risk and suggests cross-sell items. Data lives in Azure SQL: 3 tables, 10K+ transactions and 400 households.

1. CSV upload
2. Azure SQL
3. Random Forest
4. Flask API
5. Chart.js dashboard

• Random Forest models for CLV prediction and churn classification, sorting customers into Low, Medium and High retention tiers.
• Co-occurrence basket analysis over 5,000 transactions to drive cross-sell recommendations.
• Cut dashboard load time by about 80% by moving aggregation into SQL and adding in-memory caching.
• Fixed a SQL clause-ordering bug and replaced row-by-row inserts with executemany() for bulk CSV uploads.

View code on GitHub

A conversational movie assistant that answers in natural language using movies it retrieves, and declines questions that aren't about movies.

1. Question
2. MiniLM embedding
3. FAISS search
4. Gemini 1.5 Flash
5. Streamlit chat

• Started as a content-based system on the TMDB 5000 Movies dataset, using CountVectorizer (about 5,000 features) and cosine similarity, with posters, cast, director and ratings from the TMDB API.
• Rebuilt around all-MiniLM-L6-v2 embeddings and FAISS retrieval, with Gemini generating the answer from the retrieved context.

View code on GitHub

An end-to-end pipeline that turns government press releases into narrated videos in 13 languages.

1. Press release
2. Summarize
3. Translate
4. Narrate
5. Lip-sync video

• Text summarization and translation into 13 languages, gTTS narration, Wav2Lip lip-sync and video assembly with OpenCV, MoviePy and FFmpeg, behind a Streamlit frontend.
• Integrated a prebuilt GAN-based video generation model into the pipeline.
• Co-authored a peer-reviewed paper: Text-To-Video Conversion of PIB Press Releases Using Generative Adversarial Networks, IJEIMS, December 2024.

Control the cursor with your hand through a webcam: move your index finger to move the pointer, pinch to click.

1. Webcam frame
2. Hand landmarks
3. Map to screen
4. Cursor and click

View code on GitHub