Dipak B- Senior GenAI Engineer

Major Projects

Machine Learning Automation Platform – Time Series Forecasting for Agricultural Demand Planning

Built a machine learning automation platform focused on USDA and global agricultural products for time series forecasting. Implemented demand planning modules to support inventory management and strategic decision-making. Developed and fine-tuned advanced forecasting models using statistical and ML techniques, while performing multivariate analysis to uncover dependencies among products and external factors. Debugging and documented findings ensured platform stability to support transparency and knowledge sharing.

USA

Backend:

FastAPI
Keras
NumPy
Pandas
Python
Scikit-learn
TensorFlow

DevOps & Tools:

AWS
GitHub

Real-Time YouTube Query Agent – Multilingual Video Understanding with RAG and Pinecone

Designed and implemented a real-time video query assistant that enables users to ask questions about YouTube videos across languages. Developed a JavaScript browser extension to extract video context and integrated it with a Flask backend. Built APIs to fetch transcripts and generate embeddings, storing them in Pinecone for efficient retrieval. Implemented a RAG-based architecture using multilingual LLMs to deliver accurate, language-agnostic responses, tested across diverse video content for scalability and precision.

India

Frontend:

JavaScript

Backend:

Flask
Hugging Face

DevOps & Tools:

GitHub
REST APIs

Supply Chain Optimization with Reinforcement Learning – Dynamic Manufacturing Strategy Using PPO and Forecasting

Developed a reinforcement learning system using Stable Baselines PPO and Gym to optimize manufacturing rates within a supply chain context. Designed a custom simulation environment and integrated demand forecasting models to guide the RL agent’s decisions. Trained and fine-tuned the PPO model to maximize long-term profitability by adapting to changing demand patterns. Collaborated with domain experts to validate outcomes and align model behavior with real-world business objectives.

India

Frontend:

Streamlit

Backend:

Gym
Python
RLlib
Stable Baselines

DevOps & Tools:

GitHub
Linux

Healthcare Virtual Assistant – RAG-Powered Medical Query Resolution with Streamlit Interface

Architected a healthcare virtual assistant using a RAG-based pipeline powered by Hugging Face embeddings and LLaMA index to deliver accurate, real-time responses to medical queries. Preprocessed and integrated healthcare datasets to support efficient retrieval and generation. Developed an intuitive Streamlit interface for end-user interaction and validated model accuracy against real-world healthcare scenarios. Documented methodologies to ensure reproducibility and compliance with data standards.

India