Haaroon Afroz Ognawala

Coordinated research progress, experiments, and implementations across a team of four students working on computer vision and deep learning. Master's thesis research contributed to a paper accepted at the ECCV 2026 CDEL Workshop.

Co-developed and deployed a production-oriented, zero-setup RAG platform on GCP, taking primary ownership of AI engineering and implementation decisions. Designed and implemented end-to-end website ingestion, two-stage hybrid retrieval, and metadata-aware filtering; integrated retrieval workflows with Python/FastAPI backend services.

Investigated annotation variation in multi-annotator datasets, using CVAT for document-layout labeling and exploring aggregation methods that preserve disagreement.

Developed and evaluated an LLM-based company-to-tender matching system, producing structured relevance scores and justifications using expert-labeled examples. Fine-tuned and evaluated models against zero-shot baselines, enabling expert review of tender matches in seconds rather than minutes.

Built real-time advertising data pipelines and identified redundant ad spending, contributing to approximately $30K savings (10% of budget).

Analyzed and cleaned historical sales and financial data to identify operating trends, gaps, and decision-relevant business patterns.

M.Sc. in Digital Engineering with thesis submitted and defense pending.

B.Tech. in Mechanical Engineering from NMIMS University, Mumbai.