Research Intern @ Idiap Research Institute

• Proposed Concatenative Neural Speech Synthesis, a novel framework integrating Festival unit selection with WavLM SSL features to bypass supervised acoustic model training.
• Engineered a direct feature concatenation pipeline combined with kNN-Voice Conversion to enable zero-shot multi-speaker synthesis on the LibriSpeech corpus.
• Achieved +0.81 UTMOS gain in naturalness over baseline systems by leveraging HiFi-GAN vocoding and spectral smoothing in the latent space.
• Co-authored a research paper (equal contribution) titled "Concatenative Neural Speech Synthesis", submitted to Interspeech 2026.

Founding Engineer @ Novare Talent

• Architected and deployed a full-stack AI-powered talent matching platform using Next.js frontend and Python FastAPI backend with Supabase for real-time database management and authentication.
• Built intelligent AI-driven form creation and evaluation system leveraging OpenAI and Google Gemini APIs to dynamically generate customized assessments and automatically evaluate student responses with precision scoring.
• Implemented resume parsing and intelligent matching engine with cached text processing for rapid candidate-to-opportunity matching, helping companies find best-fit talent and students discover ideal opportunities in real-time.

AI & Systems Intern @ SalesAgent.ai

• Engineered a production-grade real-time multilingual speech-to-text pipeline integrating OpenAI Whisper and Sarvamai STT with hybrid fallback mechanisms for robust transcription across English and Indian languages.
• Implemented language detection and dynamic agent switching using LangDetect to automatically identify caller language (Hindi, Marathi, Tamil, etc.) and route to appropriate agent configuration with context preservation.
• Developed end-to-end voice synthesis using AI4Bharat TTS for Indian language support, enabling sales agents to respond in customer's native language with natural prosody and cultural nuance.
• Optimized streaming audio processing with chunked buffering and background VAD (Voice Activity Detection) to reduce latency by 40% while maintaining accuracy for concurrent voice agent calls.

Research Intern @ Arizona State University

• Lab: Power Electronics & Control Lab (Prof. Ayan Mallik) - focusing on advanced converter optimization and AI-driven control.
• Developed a Physics-Informed Neural Network (PINN) architecture integrating TensorFlow/Keras with differential equations to estimate parametric uncertainties (inductor saturation, parasitic resistances) in Dual Active Bridge (DAB) DC-DC converters with 92% accuracy.
• Engineered comprehensive dataset pipeline with MATLAB simulation of physics-based models, created 10k+ training samples with systematic parameter variation, and implemented data preprocessing with scikit-learn scalers for robust neural network training.
• Validated PINN predictions against hardware experimental data and comparative MATLAB-Simulink models, achieving superior generalization compared to purely data-driven baselines across unseen parameter regimes.
• Designed and implemented adaptive loss-optimal control algorithm leveraging PINN-estimated parameters to dynamically adjust switching commands, achieving 67% efficiency improvement under varying load and temperature conditions.

Indian Institute of Technology, Bombay

Dual Degree (B.Tech + M.Tech) in Electrical Engineering

Specialization: Communications and Signal Processing (CSP) | Minor: Centre for Machine Intelligence and Data Science (C-MInDS)

EPFL (Swiss Federal Institute of Technology)

Semester exchange student pursuing advanced coursework while conducting research in Automatic Audio Processing.

Arihant Public School, Kota

Pursued my senior secondary education in Science & Maths, CBSE

Secured 96.3% in CBSE Class 12th Board Examination

Little Flower School, Balangir

Pursued my primary and secondary education, ICSE

Secured 97.6% in ICSE Class 10th Board Examination