Yatharth Kapadia Email & Phone Number

Santa Clara, Ca, Us

Bloomington, Indiana, United States

Bloomington, Indiana, United States

•Evaluating and studying the behavior of large language models (LLMs).•Conducting tests on various LLMs available through Hugging Face to identify cybersecurity vulnerabilities.•Developing methods to reduce hallucinations and enhance the reliability of LLMs.

San Francisco, California, United States

Contributed to the development of an in-house fine-tuned LLM using Llama 3.1, reducing reliance on the ChatGPT API.Produced and deployed streaming APIs for STT (speech-to-text) and TTS (text-to-speech) modules using Whisper and WebSocket on an EC2 instance.Successfully optimized the entire speech-to-speech pipeline to operate with a latency of under 800ms, enhancing real-time processing capabilities.

Pune, Maharashtra, India

•Revamped the IBank platform with a Python-based multithreaded visualization tool, cutting computation times by up to 10%.•Streamlined data analytics for global hotel companies by computing 10 complex graphs, enhancing efficiency and clarity in data interpretation.•Automated repetitive tasks such as Excel charting and macros, saving over 500 hours of manual work and significantly improving productivity.•Galvanized a solid data validation system with PowerBI, salesforce and Python, running 8 essential tests to guarantee training-purpose data quality and applicability. Once verified the system produced detailed report on its own and sent it to the officers.

Mumbai, Maharashtra, India

•Spearheaded development of an innovative algorithm, designed to forecast precise time stamp at which a cricketer would throw the ball during an over. By integrating rule-based parameters and player-coordinate data, algorithm attained an impressive accuracy rate of 95%.•Reimagined an object tracking system in the Yolo model, assigning unique IDs and retaining memory for each detected object. The upgrade improved operators’ situational awareness by 90%.

United States

•Pioneered Open-Notif an application serving as a comprehensive user-company connection solution which could be accessed in form of Command-line interface. The application leveraged Pub/Sub, AWS Chalice, MongoDB and external APIs, with swift processing of 3 seconds timeframe.•Redefined NLP framework for the Ringisho app, deploying models such as profanity detection with a 95 percent accuracy rate and phrase recommendation, all hosted as FastAPI endpoints on GCP Compute Engine.•Catalyzed an advanced web-based KYC platform utilizing WebAssembly, TensorflowJS and OpenCV. User identity verification was done with Top3 accuracy of 90%, by processing data on client-side, enhancing security and efficiency.

Master Of Science - Ms, Computer Science

B.Tech, Electronics And Communications Engineering, 9.32

Education record