About
Varad Mhetar
Computer Science undergraduate with hands-on experience building Speech AI, conversational AI, and explainable machine learning applications. Experienced in developing ASR and LLM-based pipelines, implementing research-driven AI systems, and building interpretable ML models using XGBoost and SHAP. Proficient in Python, PyTorch, Hugging Face Transformers, and FastAPI, with interests in LLMs, Speech AI, and Reinforcement Learning.
Varad Mhetar
Speech AI & Machine Learning Engineer
Experience
Career & Education
work
2024 — Present
Speech AI Intern
AZMTH · Sangli, India
Developing modular Python components and end-to-end speech AI pipelines for conversational AI workflows.
- Developed modular Python components for conversational AI workflows integrating ASR and LLM-based response generation
- Built end-to-end speech AI pipelines using Automatic Speech Recognition (ASR)
- Debugged, evaluated, and optimized conversational AI pipelines, improving reliability while supporting rapid prototyping
education
2023 — 2027
B.Tech in Computer Science Engineering
D. Y. Patil College of Engineering and Technology, Kolhapur · Kolhapur, India
Pursuing Bachelor of Technology (B.Tech) in Computer Science Engineering.
- Hugging Face — LLM Fine-Tuning Certification
- Stanford Online & DeepLearning.AI — Supervised Machine Learning: Regression and Classification
Expertise
Technical Skills
The stack behind my speech AI, machine learning, and research engineering work. Select a technology for details.
01
Languages
The primitives behind every system I build.
02
AI / Machine Learning
Deep learning and interpretable modeling for real systems.
03
NLP / Speech / LLM
End-to-end systems from audio to language.
04
Engineering / Backend
Turning research prototypes into reliable services.
Philosophy
“I believe good ML systems are reproducible, interpretable, and grounded in solid engineering — where research informs implementation and real-world pipelines validate the ideas.”
Fun Facts
01
Reproduced the TurboQuant paper in Python, implementing MSE-optimal vector quantization from scratch.
02
Built an XGBoost cruise revenue forecaster achieving an R² of ~0.74.
03
Certified in Hugging Face LLM fine-tuning and Stanford's Supervised Machine Learning course.
04
Interested in LLMs, Speech AI, and Reinforcement Learning.