Project 01 · LLM tooling
A question-answering layer over tabular data that refuses to guess. The model decides what to compute; Pandas does the computing, so every number in an answer is traceable to an executed operation.
Built an AI analytical agent using LLM function calling and Pandas, supporting 10+ deterministic data operations across datasets with up to 50K+ rows.
Designed a tool-augmented reasoning architecture that combines LLM decision-making with Pandas execution engines to perform aggregations, trend analysis, anomaly detection, and data exploration with verifiable outputs.
Developed structured-response and visualization pipelines supporting 5+ chart types and validated end-to-end behavior across 75+ analytical queries, automatically producing computed insights, tables, and visualizations from user requests.