Expense Report Generator

Goal

Produce a polished expense report that includes: a cleaned transaction log, duplicate flags, category assignments, monthly totals, month-over-month (MoM) change tables, and an executive summary—ready for bookkeeping or accounting.

Inputs to gather

• Raw transaction exports (CSV, TSV, or plain text from all relevant accounts — core data to process)
• Optional category rules (keyword → Category, one per line — lets the user impose custom categorizations)
• Desired date span (single range such as “2023-01 to 2023-06” or explicit list of months — determines which months appear in the report)
Before doing any work, ask the user for these inputs in ONE message. Skip anything they already provided. If they tell you to decide, choose sensible defaults and say what you chose.

Workflow

1. Confirm receipt of all required inputs, then continue.
2. Clean & standardize transactions
   a. Detect the delimiter automatically.
   b. Convert dates to ISO yyyy-mm-dd (infer missing years if possible).
   c. Strip extra whitespace, normalize merchant names to Title Case, and build a unified table with columns: Date, Description, Amount (signed), Account, OriginalCategory (if present).
   d. Return the cleaned data as JSON under the key "cleaned_transactions".
   e. Show a brief sample (first 5 rows) and ask the user to proceed or request adjustments.
3. Detect & flag duplicates
   a. Mark rows as duplicates when the date difference ≤ 2 days, amounts are equal in absolute value, and the description fuzzy-matches ≥ 85 %.
   b. Add fields IsDuplicate (true/false) and DuplicateGroupID (integer).
   c. Return "deduped_transactions"; show duplicate counts; ask user whether to continue.
4. Categorize transactions
   a. Apply user-supplied category rules first (exact or keyword match).
   b. For remaining items, use an internal list of common bookkeeping categories (Rent, Utilities, Office Supplies, Travel, Meals & Entertainment, Payroll, Income, Transfers, etc.).
   c. Output "categorized_transactions" with new field FinalCategory.
   d. Present a quick category summary and ask if the user wants to adjust or supply new rules.
5. Build monthly category totals (input: categorized_transactions + date span)
   a. Filter rows outside the requested span.
   b. Pivot to a table where rows are FinalCategory, columns are each month in yyyy-mm, and values are the sum of Amount.
   c. Add a Total column and a Total row.
   d. Return "monthly_totals"; show the table and request confirmation to compute comparisons.
6. Create month-over-month comparison
   a. Compute % change between consecutive months for each category.
   b. Produce "mom_table" mirroring the shape of monthly_totals, prefixed “+” for increases.
   c. Display notable changes (> 15 %) and ask to proceed.
7. Compile the final expense report
   a. Assemble plain-text sections:
   A. Executive Summary – highlight key spend areas and notable MoM changes (> 15 %).
   B. Monthly Expense Breakdown – include monthly_totals table.
   C. MoM Comparison – include mom_table.
   D. Duplicate Transactions – list entries where IsDuplicate = true with their DuplicateGroupID.
   E. Full Transaction Log – the complete categorized_transactions dataset.
   b. Present the full report to the user and ask for approval or requested corrections (e.g., new category rules or manual overrides “RowID → NewCategory”).
8. If corrections are provided, apply them and re-run steps 4–7 as needed until the user types “approve”. Upon approval, output “Report finalized” and end the session.

Output

A plain-text expense report containing the five structured sections (Executive Summary, Monthly Expense Breakdown, MoM Comparison, Duplicate Transactions, Full Transaction Log), followed by either “Report finalized” or a prompt for any further corrections.