I'm Sepehr Ardebilianfard

Mostly worked for companies, learned online while attending mandatory classes.

Built the entire Wrapt platform from the ground up, creating a comprehensive SaaS application with React, Material UI, and GraphQL for the frontend, and PostgreSQL for the backend. Developed key features such as multi-user functionality, task management, and digital signatures, which facilitated the onboarding of the platform's first two customers.

Built AI products used by thousands of users, exploring product-market fit.

• Products:
• - WhatLetter: Select your language, scan, and understand. A chatbot that speaks in your native language explaining any scanned documents.
• - ImmigrationGPT: AI-powered immigration question-answering system, providing accurate and up-to-date information on US Immigration issues.
• - GetThat: A smart purchasing AI agent that offered personalized product recommendations based on customer requirements.
• - DriveHome: An AI therapist capable of maintaining context across multiple sessions, enhancing the continuity of care.

As an engineer on the Guardian team at Embark Trucks, I am responsible for developing and maintaining real-time vehicle telematics and operations systems. I have designed, implemented, and deployed a vehicle status notification API end-to-end, which allows our customers to receive timely updates on the status of their vehicles and take appropriate action when needed. My work has significantly improved the reliability and responsiveness of our telematics systems, resulting in better customer satisfaction and operational efficiency for the company.

Worked on CreditWise, a credit & fraud monitoring iOS application.

At Capital One, I designed and implemented predictive models using the Keras deep learning framework and advanced architectures such as Gated Recurrent Units (GRU) and attention mechanisms. I utilized graph embeddings of account transactions to predict monthly spending patterns and detect anomalies. My models were trained on large, complex datasets and achieved high accuracy and performance.

Developed a visual barcode generator, increasing image recognition model precision by 73% for identifying hard-to-differentiate grocery products. Implemented object detection and feature extractor models for client retailers using TensorFlow, OpenCV, and other ML libraries, achieving 98% accuracy. Improved baseline accuracy by 5% using the Adaptive Threshold algorithm to identify false positives. Successfully managed clients from pilot to production, providing technical support and staff training on data collection and validation.