About
Hi, I'm Peshal Nepal, an AI & ML Engineer who enjoys creating practical, smart systems that make things easier and more interesting. I love mixing natural language processing, computer vision, and automation to turn complex ideas into working solutions.
Over the years, I’ve worked on projects like building an AI-powered ATS with BERT, an exercise monitoring system using OpenPose and LSTM, and real-time security and virtual try-on systems using YOLOv5, FaceNet, and U-Net. I enjoy working with FastAPI, Docker, and cloud tools to deploy AI that feels seamless and reliable. Always learning, always building.
Experience
Artificial Intelligence (AI) Developer · SalesOptAI
Worked on developing a multi-assistant AI system enabling web and voice interactions using GPT-4 and GPT-5. Created an automated knowledge-base pipeline with Trieve and a custom crawler to extract, update, and manage businesses data. Developed a dynamic tool-integration framework through AgentOps Studio for custom business automation. Improved system performance and scalability through Azure-based container deployment, helping grow the platform to over 20,000 users.
Machine Learning (ML) Developer · ACID Integrations
Led a team of 5 engineers to build an AI-powered exercise monitoring system using OpenPose and LSTM for real-time exercise detection and counting. Developed a custom data collection pipeline and used Airflow to orchestrate data transfer from users to S3 and in-house servers. Improved model accuracy from 62% to 75% through dataset refinement and retraining, enabling patients to effectively follow physician-recommended exercises and share automated reports.
Artificial Intelligence (AI) Developer · Tekkon Technologies
Developed an AI-powered Applicant Tracking System (ATS) using a BERT model trained on a custom resume dataset labeled with Doccano for Named Entity Recognition (NER). Automated extraction of key resume entities and matched them with job descriptions, reducing screening time by 40%. Also built a document classification system for mining clients to categorize files by project phase and auto-fill forms, lowering review costs by 60%. Deployed scalable AI services using Docker and CI/CD on AWS ECS.
Computer Vision (CV) Engineer · EKbana Solutions
Started as a Computer Vision Intern implementing core CV algorithms in C++ (image transforms, SURF, ORB) and building simple NN/CNN from scratch. Developed a security system on Raspberry Pi with CCTV feeds using YOLOv5 for object detection and FaceNet for face detection/recognition, sending real-time alerts and improving company security by over 70%. Built a virtual try-on system using U-Net for segmentation and a conditional GAN to render garments on users, increasing customer engagement by 40%. Led weekly AI study sessions on architectures like AlexNet, VGG, RNN, and LSTM, mentoring engineers on modern CV research.
Projects
YOLO Real-Time Detection
Custom dataset curation, augmentation, training, and a real-time inference API. Monitored latency and throughput, with automatic rollbacks.
BERT Domain Adaptation
Fine-tuned BERT for classification and retrieval with evaluation dashboards and A/B tests in production.
Time-Series Forecasting
LSTM-based forecasting service with feature pipelines, monitoring, and scheduled retraining.
Time-Series Forecasting
LSTM-based forecasting service with feature pipelines, monitoring, and scheduled retraining.
Time-Series Forecasting
LSTM-based forecasting service with feature pipelines, monitoring, and scheduled retraining.
Time-Series Forecasting
LSTM-based forecasting service with feature pipelines, monitoring, and scheduled retraining.