Yasaswini dharmavarapu

Designed and implemented a PyTorch deep-learning research pipeline for music synthesis and audio pattern recognition, creating a reproducible framework adopted for ongoing faculty-supervised Music AI publications. Built automated audio feature extraction workflows (MFCCs, spectrograms) using Python/Librosa, reducing data preparation time by 50%+ and speeding experimentation cycles.

Owned end-to-end engineering for the AI Clinical Copilot, delivering ICD-10 code recommendation, drug interaction warnings, and treatment pathway suggestions using GPT-4 prompt engineering and LangChain in a production healthcare setting. Built voice-driven clinical documentation features that reduced physician documentation time by ~25 minutes per doctor using Whisper and Wav2Vec2.

Contributed to a real-time speech-to-text pipeline that transcribed physician speech with Whisper and Wav2Vec2 and converted it into structured EMR fields, eliminating manual data entry after consultations. Developed LLM-powered auto-note generation using LangChain and OpenAI prompts to transform transcriptions into formatted, clinically accurate SOAP notes, reducing documentation time by ~25 minu

Fine-tuned BERT on domain-specific employee feedback to classify sentiment and surface actionable patterns, achieving 95%+ accuracy across 500+ employee records that informed leadership strategy changes. Automated data cleaning and preprocessing with Python (Pandas, regex, custom validators), reducing manual processing time by 65%.

Pursuing an M.Sc. in Artificial Intelligence at BTU Cottbus and serving as an Academic Research Assistant building deep-learning pipelines for Music AI (music synthesis and audio pattern recognition).

Completed a B.Tech in Computer Science and Engineering with published research on bilingual Tamil–Telugu speech translation accepted at AIST-2024 (Springer).