Mithun Shivakoti

Artificial Intelligence Intern at DriveHealth

Customized and deployed HITL platforms (Argilla, Label Studio) to streamline annotation workflows and support internal ML model validation. Developed a DistilBERT-based low-latency inference model with 98% accuracy, optimized for real-time deployment in a voice AI agent, enabling accurate and safety-aligned responses during live user interactions. Isolated user operations into a standalone microservice with a dedicated database, exposing REST APIs and enabling integration via HTTPS through a gateway and internal TCP communication.

Software Development Engineer Intern at Signal X

Developed an OCR model for solving captchas, elevating daily solved instances from 70,000 to 700,000. Engineered an adverse media module with 96% accuracy for identifying adverse media related to companies. Implemented a summarization tool that condenses 30+ page risk reports into 1-page summaries, reducing analysis time from hours to minutes. Led the development of an LLM-powered chatbot for regulatory compliance and GST insights reaching 50,000 businesses.

Machine Learning Engineer Intern at TE Connectivity

Built a comprehensive model pipeline for signal integrity optimization using Gaussian Process Regression, Kernel Ridge, and Multi ElasticNet CV, reducing data transmission loss to ~1%. Organized models into a clear pipeline structure to enhance usability and clarity.

Artificial Intelligence Intern at DriveHealth

Customized HITL platforms (Argilla, Label Studio) for internal ML validation, improving labeling efficiency. Deployed DistilBERT model for voice AI with 98% accuracy and <100ms latency in real-time inference. Isolated user operations into a standalone microservice with a dedicated database, exposing REST APIs and enabling integration via HTTPS through a gateway and internal TCP communication.

Software Development Engineer Intern at Signal X

Built an OCR model to automate captcha solving (↑900%), scaling throughput from 70K to 700K daily. Engineered an adverse media classification module (96% accuracy), improving corporate risk analysis speed. Automated summarization of 30+ page reports into 1-page briefs, cutting analysis time by 98%. Deployed an LLM-powered chatbot to assist 50,000+ businesses in regulatory compliance and GST insights.

Junior Machine Learning Engineer Intern at Omdena

Engineered models to identify early autism in children with high precision, using data analytics to enhance diagnostic accuracy. Contributed to the deployment of scalable AI solutions supporting autism detection and improved healthcare outcomes.