AI-powered expense intelligence platform. Upload your company's transaction data and policy documents, then query your spending in plain English, enforce policy automatically, and surface insights through clean visualizations — all in one dashboard.

• Dashboard Overview — KPI metric cards with month-over-month changes, spend breakdown donut chart, monthly trend line, top merchants, recent transactions with manual flag support
• Talk to Data — Multi-turn natural language chat interface. Ask about spend, vendors, trends, or policy violations and get answers with inline charts and breakdown tables
• Policy Engine — Automated compliance analysis across all transactions. Flags violations (meal limits, weekend spend, foreign transactions, alcohol, tickets, duplicate splits, etc.) with severity levels and evidence
• Pre-Approvals — Paginated approval queue with AI-generated approve/review/deny recommendations per transaction
• Expense Reports — Transactions clustered by location, tag, and date into expense reports with linked policy findings
• Knowledge Setup — Provision the AI assistant, upload policy PDFs and transaction datasets, manage indexed documents

• POST /api/auth/bootstrap — Create organization and user profile on first sign-in

• POST /api/chat — Natural language query; returns answer, threadId, visualization

• GET /api/dashboard/overview — Metrics snapshot, top finding, top report
• GET /api/dashboard/approvals?page=N — Paginated approvals with recommendations
• GET /api/dashboard/policy?page=N — Paginated policy findings
• GET /api/dashboard/reports?page=N — Paginated expense report clusters
• POST /api/dashboard/policy — Execute custom policy workflow
• PATCH /api/dashboard/transactions/[id]/flag — Toggle manual flag on a transaction

• POST /api/onboarding/import — Parse and import XLSX transaction file + optional policy PDF
• POST /api/onboarding/complete — Finalize onboarding with department config

• GET /api/backboard/provision — Check assistant provisioning status
• POST /api/backboard/provision — Provision new assistant for org
• POST /api/backboard/documents — Upload and index policy/transaction documents

id, org_id, dept_id, submitted_by
posting_date, transaction_date, transaction_code
merchant_name, merchant_city, merchant_state, merchant_country, merchant_postal
description, amount, debit_credit, mcc, conversion_rate, transaction_category
flagged (boolean)
policy_status ('approved' | 'needs-approval' | 'blocked')
policy_tags (text[])
policy_last_evaluated_at
created_at

id, name, onboarding_config (jsonb), created_at

onboarding_config tracks: onboardingComplete, transactionsImported, departmentsConfigured, policyUploaded, backboardAssistantId, backboardModelName, policyDocumentIds, transactionDocumentIds

id (→ auth.users), org_id, dept_id, email, full_name
role ('admin' | 'finance' | 'manager' | 'user')
created_at

id, org_id, name, is_active, created_at

• bootstrap_organization(org_name, full_name) — Creates org + user atomically, returns org_id and role. Runs as SECURITY DEFINER.

supabase/migrations/
  001_initial.sql                              — core transactions table + RLS
  20260404143000_bootstrap_organization.sql   — org bootstrap function
  20260404173000_add_policy_status_to_transactions.sql
  20260404180000_add_flagged_to_transactions.sql

8 built-in rules, each producing typed findings with severity:

Custom policy workflows can be designed in the Policy Engine using a node/edge graph builder. Supports a DSL with hasTag(), missing(), withinHours(), withinDays(), sum(), count(), and field comparisons.

Transactions are auto-tagged by merchant name and MCC code:

software · shipping · fuel · travel · lodging · meals · parking · fees · alcohol · tickets · other

1. User uploads XLSX via onboarding wizard
2. ExcelJS parses file with dynamic column header normalization
3. Departments auto-detected and created in Supabase
4. Transactions inserted in 500-row chunks
5. Optional policy PDF uploaded to AI assistant and polled until indexed

1. User query → resolveQueryState() parses intent (focusTag, compareTag, period, mode)
2. buildVisualization() generates deterministic analysis
3. If assistant is provisioned, sends query + analysis JSON + policy context to LLM
4. LLM returns narrative answer; fallback to local narrative if unavailable
5. threadId persisted for multi-turn conversation continuity

Transactions grouped by location + tag + 4-day date buckets. Groups with ≥ 2 transactions or a single transaction > $250 become expense reports. Each report includes linked policy findings and an AI approval recommendation.

src/
  app/
    api/                  # All API route handlers
    dashboard/            # Dashboard pages
    auth/                 # Login / callback
    onboarding/           # Onboarding wizard
  components/
    app/                  # Feature components
    ui/                   # Base UI primitives (shadcn + custom)
  lib/
    data.ts               # Transactions, snapshot, visualization
    policy.ts             # Policy engine (~1,238 lines)
    policy-rules.ts       # Rule definitions and DSL
    dashboard.ts          # Paginated payload builders
    reports.ts            # Expense report clustering
    organization.ts       # Org context, roles, onboarding config
    backboard.ts          # LLM assistant API integration
    format.ts             # Currency, number, date formatters
    file-types.ts         # MIME type detection helpers
    supabase/             # Browser, server, and middleware clients
  hooks/
    use-file-upload.ts    # Dropzone + upload state
    use-json-query.ts     # Fetch + cache with refetch
supabase/
  migrations/             # SQL migration files
training/
  model/                  # Python ML anomaly detection models

# Supabase (public)
NEXT_PUBLIC_SUPABASE_URL=
NEXT_PUBLIC_SUPABASE_PUBLISHABLE_KEY=

# Backend secrets
NEXT_BACKBOARD_API_KEY=
NEXT_BACKBOARD_ASSISTANT_ID=
NEXT_BACKBOARD_MODEL_NAME=
NEXT_BACKBOARD_POLICY_DOCUMENT_ID=

npm install
npm run dev

Open http://localhost:3000.

Apply database migrations via the Supabase dashboard SQL editor or CLI before first use.