Selected work

Project 02 / Agentic AI

TalkToData

From a spreadsheet to a useful answer.

Explore source on GitHub
Live demo

Overview

A closer look
at the system.

A full-stack AI analytics application that turns questions about CSV and Excel files into computed insights and charts. Includes authenticated uploads, persistent conversations, usage tracking, and chart storage.

Engineering focus
Full-stack + applied AI
What this demonstrates
A Next.js interface, FastAPI backend, authentication, persistent data, and an analytics agent.
Project context
Full-stack & AI development
Technology
OpenAI Agents SDK / Gemini / Next.js / FastAPI / Supabase
System study / 02Conceptual flow
  1. 01
    Upload spreadsheetNext.js
  2. 02
    Ask a questionFastAPI
  3. 03
    Run analysis toolsOpenAI Agents SDK
  4. 04
    Return charts & answerspandas · Supabase
Agentic AI

01 / The problem

Working with a spreadsheet should not require writing a new analysis script for every question.

02 / Implementation & design

The OpenAI Agents SDK orchestrates a ReAct loop over file, analysis, and visualization tools. Inference uses Gemini through Google’s OpenAI-compatible endpoint, with a configurable model. Next.js provides the interface, FastAPI handles the backend, and Supabase supports authentication, persistence, and storage.

03 / Engineering challenge

Thirteen agent tools cover statistics, filtering, grouped aggregation, correlation, and chart generation. Authenticated uploads, persistent conversations, usage tracking, and stored charts support the application around those tools.

Results & capabilities

13 agent tools

Statistics, filtering, aggregation, correlation & charts

Gemini + OpenAI Agents SDK · Dockerized deployment