Hire Rohith - Fine-Tuning Engineer

● Automated end-to-end reporting workflows in Python and SQL for the SGC back-office group, cutting manual effort on weekly operational scorecards by ~60% and freeing analyst time for higher-value analysis. ● Built interactive Power BI and Tableau dashboards tracking 8+ KPIs across productivity, service levels, and capacity utilization, giving senior stakeholders same-day operational visibility in place of lagged weekly reports. ● Engineered ETL pipelines to ingest, normalize, and aggregate data from 5+ heterogeneous sources (databases, spreadsheets, external feeds) into a unified analytics layer, eliminating manual data stitching across teams. ● Applied forecasting and variance analysis in Python (Pandas, NumPy, scikit-learn) to predict operational trends, improving resource-allocation accuracy by an estimated 20%. Partnered with actuarial and operations teams to translate ambiguous business requirements into structured data solutions, delivering weekly status reports and ad-hoc analysis on a 24-hour turnaround.

● Collected, cleaned, and standardized data from 6+ source systems using Python and SQL, cutting data-prep time by ~40% through reusable transformation scripts. ● Built 4+ interactive Power BI and Tableau dashboards to visualize business KPIs, enabling non-technical clients to self-serve insights and removing a recurring analyst bottleneck. ● Applied regression, trend analysis, and cohort segmentation across operational datasets to surface patterns, delivering 3 actionable process-improvement recommendations to stakeholders. ● Automated recurring report generation with Python, eliminating ~5 hours of manual effort per week across the analytics team. ● Maintained a 0-error rate on client-facing deliverables through a structured QA validation checklist, safeguarding data accuracy and stakeholder trust.

● First-authored a peer-reviewed poster on a mental-health NLP pipeline, engineering a Python web-scraping and preprocessing workflow that curated a 2,750-post ReachOut corpus (3.38M+ tokens) across 8 disorder categories, with lemmatization, HTML/emoji cleaning, and PII anonymization to deliver an ethically compliant, model-ready dataset. ● Trained LSTM and BiLSTM classifiers and fused them into an ensemble for multi-label detection of coexisting disorders, achieving up to 83.33% post-classification accuracy and validating the curated corpus for downstream use. ● Developed a BERT-based summarization module to distill lengthy forum threads into concise summaries, reaching ~54% ROUGE-1/ROUGE-L F1 and enabling faster triage and navigation of high-risk posts. ● Annotated 16,280 clinical EHR sentences for suicide ideation (SI) and attempt (SA) labels and co-designed CMTN, a novel Convolutional Multi-Level Transformer combining CNN, transformer, and attention layers to capture local and global text dependencies for suicidal-behavior prediction. ● Benchmarked CMTN against state-of-the-art and baseline models across BERT, BioBERT, and PubMedBERT embeddings, attaining 0.97 precision on the SA class and leading SI recall/F1 (0.56/0.52), and open-sourced the data-alignment and annotation-extraction algorithms to accelerate reproducible suicide-prevention research. ● Built a Python web scraper to extract metadata, ratings, and reviews from 500+ Google Play mental health apps, feeding a structured analytics pipeline that surfaced market and sentiment trends.