Pravesh Budhathoki

I’m an AI/MLEngineer with 5+ years of experience building intelligent systems across computer vision, document intelligence, conversational AI, and backend platform engineering. I focus on designing scalable AI architectures and deploying production-grade machine learning solutions in domains like tax, compliance, KYC, OCR, and analytics.

At IFC (Investor Friendly CPA), I lead AI and backend engineering initiatives for a tax-domain platform centered on conversational intelligence, document automation, compliance workflows, and knowledge-driven systems. I’ve built tax-domain conversational AI workflows using intent routing and response synthesis, and developed OCR + LLM pipelines for IRS form extraction with confidence-based review systems.

I’ve also implemented semantic retrieval systems using embeddings and vector search for tax knowledge bases, along with immutable audit trails for compliance-sensitive workflows. To make these AI services resilient and scalable, I scaled async processing with Celery, Redis, PostgreSQL, and Dockerized workers—adding caching, locking, and idempotent job patterns.

Previously at Treeleaf Technologies, I advanced core computer vision and document automation capabilities, improving license plate detection accuracy from 92% to 95% and Nepali plate recognition from 97.8% to 99% across 19 classes. I built KYC verification systems for Nepali IDs using OCR and face matching, engineered real-time face attendance pipelines with vector search at scale, and developed a YOLOv7-based monkey detection model (85% recall, 82% precision).