Case Studies

Below are select projects demonstrating how I’ve applied AI, ML, and software engineering to solve real-world problems in finance, healthcare, and infrastructure.

🧠 LLM Evaluation for Health Recommendations

Built a framework to assess the reliability of AI-generated health guidance.

• Challenge: Evaluate whether GPT-based responses aligned with medical expectations

• Solution: Designed structured prompt templates and applied a custom scoring rubric

• Result: Delivered targeted feedback for improving factual accuracy in model outputs

Technologies: OpenAI GPT, Prompt Engineering, Evaluation Metrics, Numpy, Pandas

🧾 Remittance Processing with Machine Learning

Automated remittance processing using text classification techniques.

• Challenge: Improve document matching accuracy in remittance workflows

• Solution: Developed an NLP pipeline with SVM classifiers

• Result: Increased match rates and reduced manual effort

Technologies: Python, NLP, Support Vector Machines, AWS, C#

📩 Lockbox Ingestion Agent

Developed a configurable parser for financial lockbox data ingestion.

• Challenge: Handle complex and varied file formats for ingestion

• Solution: Built a regex-based parser to extract relevant fields from raw files

• Result: Enabled a successful product rollout by streamlining backend data integration

Technologies: Python, Regex, Automation, CSV/Flat File Parsing

📊 LLM Benchmarking Framework

Designed tools to compare performance across multiple LLM frameworks.

• Challenge: Assess different LLMs on tasks like QA and structured prediction

• Solution: Created benchmarks including crop yield prediction and scientific QA

• Result: Enabled informed selection of optimal LLM for downstream use

Technologies: LLM APIs, Groq, JSONL Datasets, Metric Design

🏠 3D Roof Reconstruction Using Deep Learning

Built a multi-task deep learning model to estimate roof geometry.

• Challenge: Extract azimuth, tilt, height, and perimeter from aerial imagery and point clouds

• Solution: Integrated EfficientNet and ResNet-based architectures into a shared model pipeline

• Result: Delivered an end-to-end tool for automated roof geometry estimation

Technologies: TensorFlow, EfficientNet, ResNet50, Point Cloud Processing, PyVista