Data Analyst

ai-agent
Data Analyst

Table of Contents

This is a list of all the sections in this post. Click on any of them to jump to that section.

Conversational Data Analyst

Upload a CSV or Excel file, ask questions in natural language, get back stats, charts, dashboards, and reports.

Reloading a saved dataset and generating a histogram

Scatter plot generation with query usage counter

Features

  • CSV and Excel file ingestion with type inference and cleaning
  • Descriptive analytics: profiling, summary stats, groupby, crosstab, outlier detection
  • Interactive Plotly charts (HTML/JSON) and static matplotlib/seaborn charts (PNG)
  • Self-contained HTML dashboards and PDF reports
  • Multi-user support with Supabase auth, per-user rate limiting, and dataset persistence
  • Chat commands for quick file management (/upload, /datasets, /saved, /reload)

Architecture

Built on DeepAgent with a multi-agent orchestration pattern. An orchestrator agent coordinates two specialized subagents:

  • Stats Researcher - Descriptive analytics. Profiling, summary stats, groupby, crosstab, outlier flags.
  • Chart Maker - Visualization. Chooses between Plotly (interactive) or matplotlib/seaborn (static) based on the request.

12 custom tools registered on the harness - loaders, query runners, chart generators, dashboard builders. Each subagent gets access to the tools relevant to its role. The architecture is designed to scale - new subagents can be added without changing the orchestrator logic.

Backend

  • LocalShellBackend - Sandboxed virtual filesystem for local/dev use.
  • CompositeBackend - Routes between StateBackend (ephemeral runtime) and StoreBackend (persistent cross-session memory via LangGraph Store).
  • PostgresStore (production, Supabase) or InMemoryStore (dev fallback) for long-term memory.
  • Conversation checkpointing with LangGraph MemorySaver for mid-conversation resume.
  • AGENTS.md memory paths for persistent subagent context.

Upcoming

  • Data cleaning and preprocessing through conversation
  • Time series forecasting and predictive analytics
  • Data manipulation, transformation, and export

Stack

Chainlit, DeepAgent, LangGraph, Supabase, Plotly, Jinja2, Docker Compose.