Experience

Proposed and built a ranking system using semantic retrieval to automate sustainability analysis, reducing manual review time 70%.

Developed a FastAPI backend with 8 endpoints converting 6,900+ analyst decisions into auditable training data for supervised filtering.

Engineered a 4-stage filtering pipeline with transformer embeddings and Pinecone vector search, cutting preprocessing time 30%.

Shipped a containerized review platform with Docker and a zero-dependency query cache, reducing redundant API calls 95%.

Mentored an intern via code reviews and architecture documentation; designed modular components for independent feature development.

Developed a 3-pipeline multimodal platform (speech, vision, transaction data) for borrower risk modeling, increasing data processing throughput 40%.

Transformed multilingual client interviews into structured credit signals by fine-tuning Whisper-small on regional speech data, improving transcription accuracy 31.5%.

Implemented a computer vision pipeline using OWLv2 prompt-based object detection with non-maximum suppression, automating asset verification across 10,000+ images.

Built a real-time price prediction pipeline for incomplete inventory data, improving catalog completeness 35%.

Developed anomaly detection dashboards (Tableau) and enforced data integrity via SQL-based validation.

Automated an API validation suite of 200+ tests across load-balanced AWS EC2 environments, resolving 15+ critical vulnerabilities.

Optimized system response time 30% (5s→3.5s) by redesigning request handling workflows to reduce performance bottlenecks.

Implemented user authentication and session management with MySQL-backed credential validation, strengthening secure access.