Harish Manoharan
Software engineer at Purgo AI, working on its data-engineering agent and the evals behind it.
I joined Purgo as an intern in June 2025 and went full-time in June 2026. I work across our LangGraph agent for data engineering: how it analyzes a ticket, finds the tables it needs, drafts a design, and writes and reviews Databricks and dbt code. I also build the evals and benchmarks we use to decide whether a change helped. Before Purgo, I built a text-to-SQL agent at Digitus and forecasting models at PwC.
Featured
An agent that writes data-engineering code
How it finds tables, drafts code and checks its own output, each change measured on benchmarks built from real tickets. Following dbt lineage cut the tables it missed by about 70%.
Testing LLMs for regulated use
Qualification tests for models and Databricks platforms, with evidence an auditor can read.
a11y-stem
Turning equations in STEM lecture PDFs into MathML a screen reader can speak. Private until the beta.
More work
Text-to-SQL agent
Schema discovery plus retries driven by execution errors.
Execution accuracy 52% to 76% on an internal 500-query set
Finding behaviors in mouse videos without labels
Segmented pose-tracking video of mice into behaviors with a bidirectional RNN autoencoder and clustering, for genetic analysis. Advised by Balaraman Ravindran and Vivek Kumar.