Work summary for AI-Ml Projects

A brief summary of my work:

Employee Productivity Prediction System

This AI-powered system predicts employee productivity trends across projects and months, based on 14 years of time log data. It was designed to assist businesses in strategic planning by providing insights into employee performance and project efficiency. I led the development of this predictive system, which involved simulating 9 lakh+ data points for 200 employees working on 50 different projects. We used time-series models like Linear Regression, SARIMAX, Prophet, and LSTM, as well as a hybrid LSTM + Prophet architecture for more accurate future predictions. Additionally, I implemented interactive data visualizations to clearly communicate the results to non-technical stakeholders.

Tech Stack:

• Python, Pandas, Matplotlib, Statsmodels, Facebook Prophet, TensorFlow/Keras, Seaborn

Role: Project Lead

• Led the modeling process, including data simulation, feature engineering, and the development of multiple forecasting models.
• Applied time-series techniques to predict future trends, incorporating model tuning and evaluation for optimal performance.
• Built data visualizations to facilitate decision-making and ensure clear communication of model results.

Natural Language to SQL Forecasting Assistant

This system converts natural language queries into executable SQL queries for time tracking and forecasting applications. I developed this solution using a fine-tuned T5 transformer model, which was trained on over 100 prompt-to-SQL mappings to interface with a PostgreSQL database containing time logs, project metadata, and forecast tables. The system enables the user to query the database in plain English, supports complex SQL features like cross-table joins, aggregations, and time-based analysis, and provides predictive answers based on ML-generated forecast data.

Tech Stack:

• Python, Hugging Face Transformers (T5), spaCy, SQLAlchemy, PostgreSQL, Pandas, Datasets

Role: Full Stack Developer

• Led the design and development of the NLP pipeline to convert user queries into SQL.
• Fine-tuned a T5 Transformer model to ensure schema-aware and context-specific SQL generation.
• Integrated SQL generation with forecasting tables to provide predictive results directly through SQL queries.

Multilingual FAQ Retrieval Chatbot using RAG

This multilingual FAQ retrieval system is designed to answer user questions in both English and Japanese by leveraging a Retrieval-Augmented Generation (RAG) approach. The system processes FAQ PDFs by chunking the content, indexing it with multilingual embeddings, and retrieving relevant sections for answering queries. I integrated Groq’s LLaMA 3.3 70B model for high-quality, context-sensitive responses and implemented a feedback loop to allow users to interactively ask questions. The system supports both languages, ensuring answers are given in the appropriate language, and uses advanced filtering to improve query relevance.

Tech Stack:

• Python, LangDetect, Hugging Face Sentence Transformers, LlamaIndex, ChromaDB, Groq LLM

Role: AI/ML Developer

• Developed and fine-tuned multilingual embeddings for the retrieval process.
• Implemented RAG for context-aware question answering and integrated the LLaMA model for natural language generation.
• Managed the development of an interactive CLI for iterative querying and satisfaction feedback.