Om M. Patel

Jun 2026 - Aug 2026

• Integrated vision APIs (with routing layer) with Next.js for extracting data to reduce manual work by 64%.
• Built an eval layer to measure extraction quality and invoke HITL validation and POS reconciliation on the user dashboard.
• Designed Supabase PostgreSQL-backed data storage, with location-based access and private media retention for stock audit.
• Automated email ingestion pipeline to normalize POS sales reports (EML, CSV, PDF), avoiding paid access to POS APIs.
• Shipped via AI coding agents: Playwright e2e tests, CI checks and Sentry monitoring in production to support reliable releases.

Dec 2024 - Aug 2025

• Built Python ETL pipelines to parse API logs into a PostgreSQL analytics database, serving dashboard endpoints in under 1.5 s.
• Refactored a Flask monolith into Dockerized microservices with Redis caching, cutting route latency by 30-40% (1 s to 600 ms).
• Coordinated with backend and ML teams to monitor API behavior and support iterative post-deployment improvements.

Mar 2024 - Jul 2024

• Developed geocoding APIs with Flask and GeoPandas to reduce dependency on the Google Maps API, saving about 350 USD monthly.
• Scraped and pre-processed 20k+ addresses to create training datasets, producing versioned datasets for ML training.
• Built MongoDB-backed credit and Paystack payment flows with rate limiting and CSRF protection. Shipped with Docker on Render.

Sep 2025 - Aug 2026

GPA: 3.77/4

Vector Scholarship in AI, Vector Institute ($17,500 merit award) · Supervisor: Dr. Aijun An

Relevant coursework: Probabilistic Models and Machine Learning, Machine Learning Theory, Data Mining, Fairness and Bias in AI, Data Analytics and Visualization, Introduction to Robotics

Nov 2021 - May 2025

GPA: 3.47/4

Relevant coursework: Computer Vision, Digital Image Processing, Pattern Recognition, Information Retrieval, Natural Language Processing, Big Data Analytics

Mar 2026 - Aug 2026

• Built a Gujarati sentiment-analysis pipeline across translated SST-2 and native GSAC data to evaluate polarity preservation.
• Achieved weighted F1 of 0.804 with a MuRIL-based model and a 2,908-entry sentiment lexicon on a held-out GSAC test split.
• Kept adaptation and test data disjoint and analyzed polarity shifts, negation errors, mistranslations, and social-text noise for the manuscript.

Sep 2025 - Dec 2025

• Fine-tuned Llama 3.1 8B with 4-bit QLoRA (Unsloth) on an NVIDIA RTX A6000, in a graph-augmented pipeline that converts each instance into a disjunctive graph (PyTorch Geometric) serialized into the prompt.
• Pre-processed the noisy STARJOB dataset (~130k to 9,525 usable instances), cutting fine-tuning runtime from 70h to 11.5h.

Sep 2024 - Jul 2025

• Surveyed 90+ automated sleep staging studies and organized signal modalities (EEG/ECG/EOG/EMG) and PSG data representations into an end-to-end taxonomy used for system design.
• Consolidated model families and evaluation practice (classical ML → deep learning) with reporting guidance around leakage, imbalance, and cross-subject generalization.