AboutProjectsSkillsEducationGet in Touch
Back to projects
Project · Live Demo

AR Insights Dashboard

A synthetic accounts receivable portfolio for a fictional B2B company. Ask in plain English about total exposure, aging, collection trends, or any specific customer. The assistant picks one or more analytical functions, runs them, and answers in a few sentences with inline charts and a full audit trail of the data behind every number.

Pattern: agentic tool-use loopFunctions: 6 vetted analytical primitivesModel: Claude Haiku 4.5Charts: inline SVG, no library
Total open AR
$6,456,100
24 customers · 43 open invoices
% past due
52.8%
$3,411,400 of total
Avg days overdue
34d
Across past-due balances
Largest exposure
DriftWood Hotels Group
$841,000 · 105% of limit
AR Insights · Aurelia Industries

Ask a question about Aurelia's accounts receivable portfolio. The assistant calls one or more analytical functions over the dataset, then writes the answer in plain English. Every number is traceable back to the source rows.

How it works

  1. 01
    Question arrives
    A finance lead types a question - everyday English, no SQL. The recent conversation history travels with it so follow-up questions work naturally.
  2. 02
    Plan the tool calls
    Claude sees the question and a fixed catalogue of six analytical functions, each with a strict input schema. It picks one or chains a sequence - say, a customer ranking followed by that customer's detail.
  3. 03
    Execute against the dataset
    Each function is pure deterministic code over the in-memory ledger. It returns structured records, a summary line, and (where useful) a chart specification the front end can render.
  4. 04
    Synthesise and cite
    Claude writes a two-to-four sentence answer using the tools' output. The UI renders the charts inline and opens an audit panel listing every function call, every argument, and every record returned.

How the tool could be extended