MOUNIKA BATTU

Manufacturing Product Chatbots

(September 2024 – Present)
Goal: Developed a Generative AI-powered chatbot system enabling natural language interaction with manufacturing databases. The chatbot converts user queries into SQL, delivering real-time, contextualized responses.

Highlights:

• Natural Language to SQL Conversion: Translated user queries into structured SQL commands for seamless data access.
• Contextual Responses: Implemented contextual awareness in replies using LLMs for enhanced understanding.
• Azure Embeddings Index: Leveraged Azure Cognitive Search Services for efficient semantic retrieval.
• Prompt Engineering: Crafted optimized prompts to maximize relevance and accuracy of model outputs.
• Deployment: Delivered a production-ready UI using Django and deployed backend with Docker containers.

Tools & Skills:
LLMs, Django, Python, LangSmith, Azure Cognitive Search Services, Docker

(April 2024 – Sep 2024)
Goal: Developed a Generative AI-powered virtual sales agent that interacts with potential customers, mimicking a human sales executive to drive engagement and lead conversion.

Highlights:

• Conversational Sales Agent: Designed the bot to simulate natural, sales-driven conversations to engage customers and qualify leads.
• Lead Capture & Follow-Up: Enabled the bot to collect customer details (e.g., name, email, interest) and automatically schedule meetings or send invitations via email.
• LLM-Powered Interactions: Utilized Azure OpenAI for generating persuasive, human-like sales dialogues.
• Semantic Search Integration: Implemented Azure Search Index for retrieving relevant product/service content to enhance responses.
• Seamless Experience: Ensured smooth transition from conversation to email scheduling, enhancing customer experience and reducing manual follow-up.

Tools & Skills:
Azure OpenAI, Azure Cognitive Search, Python, LangChain, Email Automation

(August 2023 – Feb 2024)
Goal: Developed a Generative AI-based educational assistant to enable students and faculty to interact with lesson materials through natural language queries. The chatbot interprets queries and retrieves precise answers from academic PDFs.

Highlights:

• Contextual QA over Lesson PDFs: Designed a retrieval-based QA system to allow students to query academic content intuitively.
• Vector Store Integration: Employed ChromaDB and PGVectorDB to store and retrieve document embeddings efficiently.
• LLM Integration: Used OpenAI models to generate concise, context-aware responses from lesson content.
• Cloud Storage: Leveraged AWS S3 for storing and accessing educational documents in a scalable and secure manner.
• User-Centric Design: Focused on ease of use for educational settings, improving student access to learning material.

Tools & Skills:
ChromaDB, PGVectorDB, OpenAI, AWS S3, Python, LangChain, LLM-based Retrieval QA

QIPP Product

(June 2022 – February 2023)
Goal: Designed and built a production-grade SaaS tool for inventory and procurement optimization in manufacturing.

Highlights:

• Modeling: Applied statistical and time series methods for demand and price forecasting.
• Techniques: Used clustering and probabilistic models for supplier ranking and inventory planning.
• Optimization: Built linear programming models for optimized order schedules and cost savings.

Tools & Skills:
Spark, Hive, Hadoop, SQL, Airflow, Logging, Git, Postman, Flask REST APIs, Azure

(Defect Detection & Flight Simulation Visualization)

Overview: Delivered solutions for defect detection in manufacturing and visual analytics in aerospace simulations.

Steel Defect Detection:

• Built image pipelines for assembly line defects using feature extraction and masking.
• Used deep learning models for binary classification and image segmentation.
• Showcased results through an interactive Streamlit application.

3D Graphical Overlay Tool:

• Developed a tool to visualize data points over 3D flight simulation images.
• Applied CV preprocessing (e.g., axis detection, line detection) and created overlays using automated plot digitizers.
• Enabled better interpretation through dynamic visualization.

Technologies Used:
Keras, TensorFlow, PyTorch, Python, Streamlit, OpenCV, Matplotlib, Seaborn

(June 2019 – May 2022)
Goal: Developed a predictive model to forecast key operational metrics — Temperature in the boiler room, Flow Rate, and Turbine Pressure — to support proactive control in power plant operations.

Highlights:

• Data Collection: Processed one year of second-wise sensor data consisting of over 200 variables (tags).
• Feature Selection: Identified top influencing variables for each target parameter using correlation and dimensionality reduction techniques.
• Forecasting with ARIMA: Analyzed trends (seasonality, cyclicity, irregularity), applied smoothing to achieve stationarity, and developed ARIMA models to forecast influencing variables at 10-minute intervals.
• Target Prediction: Used forecasted inputs in regression models to predict temperature, flow rate, and pressure with 89% accuracy (RMSE-based).
• Trend Analysis: Incorporated time series analysis using ACF and PACF to understand temporal dependencies in the system.

Analytical Tools:
Python, PostgreSQL, Excel

Techniques Applied:
Univariate Analysis, Correlation, PCA, Trend Analysis, ACF & PACF, ARIMA, Regression

• Performed statistical analysis of survey data and conducted sentiment classification using text mining techniques.
• Built foundational skills in data science, analytics, and reporting.