Kaushik Burra

Financial Prediction Service (Sep 2023 – Nov 2024)

• Objective: Developed an AI solution to detect "bad" SSIs using historical trade data as ground truth.
• Data Engineering: Engineered pipelines for labeled SSI data, preprocessing and transforming raw data into scalable, high-value features for ML models.
• Model Development: Implemented Py Torch deep learning models, Multinomial Naive Bayes, and XGBoost; created ensemble models combining these approaches.
• Experiment Tracking: Utilized Weights & Biases for comprehensive model tracking and analysis.
• Deployment: Built prediction REST services, regression-tested, and deployed to QA/Production using CI/CD.
• Impact: Improved proactive SSI detection, streamlining trade processes.
• Tools: PyTorch Lightning, Python, Airflow, SQL, Git

QIPP Product (June 2022 – Aug 2023)

• Goal: Developed an end-to-end SaaS tool for cost optimization and order scheduling for manufacturing.
• Modeling: Applied statistical and time series analysis for demand and price forecasting; designed linear programming models for order optimization.
• Techniques: Implemented Naive Bayes and clustering for supplier ranking and inventory management.
• Tools: Spark, Hive, Hadoop, SQL, Airflow, Flask APIs, Azure.

GENERATIVE AI Use Cases

Financial Document Parsing Solution

• Developed a comprehensive pipeline using multiple parsing tools (LLamaparse, Unstructured, PyMuPDF) to extract tables from financial PDFs.
• Integrated LLM-generated models, such as GPT-4 Turbo, to enhance data parsing and interpretation.
• Implemented Retrieval-Augmented Generation (RAG) with ChromaDB for efficient data retrieval and processing.
• Used retrievers and chains for structured question-answer workflows, leveraging LangSmith to evaluate and refine LLM-generated responses.
• Enhanced accuracy and efficiency in parsing complex financial data, supporting advanced data extraction and analysis.
  Tools: langchain, llamaindex, Hugging Face, langsmith

COMPUTER VISION USE CASE
Computer Vision Use Case:

Overview: Developed end-to-end solutions for defect detection in manufacturing and enhanced data visualization for flight simulations.

Steel Defect Detection:

Built a pipeline for identifying defects on assembly lines using feature extraction and image processing techniques (e.g., RLE, image masking).
Implemented binary classification of images using deep learning CNN models (ResNet50) and applied image segmentation to classify defects.
Developed a Streamlit application to showcase the entire defect detection pipeline.

3D Graphical Overlay Tool:

Created a tool for plotting data points on 3D graphical images in a flight simulation domain.
Employed computer vision preprocessing techniques (line detection, axis detection) and utilized a plot digitizer to extract data points.
Developed an automated overlay tool to visually represent data points for better user understanding.
Technologies Used: Keras, TensorFlow, PyTorch, Python, Streamlit, OpenCV, Matplotlib, Seaborn.

• From Fulfilment-to-Fulfilment Centres, scheduled Ad hoc for the nominal trucks and cancelled the ones that are not required based on predictions and ensured on time delivery to customers.
• Using Tableau and SQL, worked on queries pulling middle mile logistical data.
• Oversaw client communications, managed record tracking and data communication activities.
• Led shipping and delivery quality control by managing downtime resulting in large increase in revenue.
• Oversaw every phase of supply chain, from purchase order to delivery to invoicing, targeting100% end-user satisfaction.

•Assisted with research and gained extensive knowledge of Pratham Education focusing on improving the quality of education for underprivileged children across India.
•Furthermore, built a project roadmap, analysed the ASER (Annual Status of Education Report) data over the last one year and analysed factors contributing to low quality in the education System.
•Finally discovered Private School enrolment catching up is huge. Though most children (age 5- 12) continue to get their education from Governments schools, private school enrolment is increasing despite high costs.

Tools used: Excel and Tableau