Bhuvan Ummidisetti

At SPIRE Lab, Dept. of Electrical Engineering, IISc Bangalore, I developed a Transformer-CTC keyword spotting and forced alignment system for SWIENS, achieving ~93% keyword classification accuracy with word-level timestamps. I also built an end-to-end speech pipeline covering audio preprocessing, feature extraction, transcript tokenization, CTC training and decoding, and evaluation.

During the on-site phase, I fine-tuned Wav2Vec2-XLS-R-300M with a character-level CTC head on 37,136 speech clips. I built an evaluation pipeline against human-corrected annotations that achieved 2.11% WER and 1.74% CER.

For my Olist e-commerce BI case study, I built an analytics workflow across a 1.55M-row dataset, designed a PostgreSQL star schema, and cross-validated KPIs in SQL and Pandas. I delivered a four-page Power BI report and analyzed revenue and customer behavior.

In my Transaction Risk Observatory project, I engineered an ELT pipeline and PostgreSQL star schema for transaction data, then developed explainable SQL fraud signals and a risk score. I also built KPI views and a PL/pgSQL data-quality procedure with 98 checks.