
Data Analyst
Raga AI · Sep 2023 – Oct 2025
- Developed a platform to manage and optimize LLM projects with features like dataset and prompt management, evaluation tools, synthetic data generation, and guardrails.
- Integrated agent tracing to monitor LLM behavior, tool usage, user interactions, and performance for improved transparency and debugging.
- Built internal analytics views using Python + Plotly/Streamlit-style interactive dashboards to visualize model performance, evaluation metrics, and usage trends for stakeholders.
- Performed systematic quality analysis on image datasets used for facial expression recognition and human pose estimation tasks (FER, COCO Keypoints).
- Identified and quantified data quality risks such as duplicate records, inconsistent resolution, extreme lighting conditions, motion blur, and class imbalance impacting downstream model performance.
- Designed interactive EDA dashboards using Python + Plotly/Streamlit to explore label distributions, class imbalance, and coverage gaps across categories.
- Designed trend and diagnostic graphs to compare image quality metrics (resolution, exposure, blur) across classes, enabling faster identification of bias and data skew.
- Translated insights from visual analysis into actionable business recommendations, including data sampling strategies, outlier handling, and imbalance mitigation to improve dataset reliability and analytical outcomes.
- Developed and implemented custom evaluation metrics for clients, including context relevancy, response correctness, and response completeness.
- Conducted benchmarking of hallucination and false refusal detection using industry-standard datasets (RagTruth, HaluEval), comparing evaluation outputs across Ragas (76%) and Azure (78%) frameworks, with final agreement accuracy of 86% against reference labels.
- Collaborated with cross-functional teams to align metric definitions with client goals and model evaluation needs.



