Brim Financial × MPC Hacks. An AI-powered expense-intelligence platform for SMB credit-card spending - built on real (anonymized) company-card transaction data and the real Brim expense policy. Make the data talk.

A non-technical finance manager can chat with their company's spend in plain English, manage a digitized expense policy and triage violations, run an AI pre-approval queue, and generate review-ready expense reports - all grounded in real numbers.

1. Talk to Your Data (Ask AI, floating on every page) - Agentic conversational analytics. Google Gemini runs a function-calling loop over 5 read-only query tools (aggregate, time-series, top-merchants, list, compare), auto-renders the right viz (bar/line/pie/table/stat), and handles multi-turn follow-ups ("now just Texas, monthly" reuses prior filters) - never inventing numbers.

2. Policy Compliance Engine (Governance → Violations) - Six rules digitized from the real Brim policy ($50 pre-auth, no traffic/parking tickets, alcohol restriction, tips, etc.). Auto-scans transactions, detects split-charge evasion (same card+merchant+day crossing a threshold), surfaces repeat offenders, and applies an AI contextual severity pass - distinguishing legitimate same-day batching from genuine threshold-ducking.

3. AI Pre-Approval Workflow (Workflow → Approvals) - Each request shows the card's spend history, prior-vendor count, and category budget status, plus a Gemini approve / deny / review recommendation citing the real numbers. The approver decides once; decisions persist.

4. Automated Expense Reports (Workflow → Reports) - Transactions auto-grouped by location + month with category breakdowns, line items, policy-flag counts, and AI-written CFO summaries, ready for one-click approval.

• Anomaly & fraud - duplicate/recurring charges, round-number patterns, outliers (plus the settlement-vs-spend context flag).
• Vendor consolidation - fragmented spend (e.g. fuel across 218 vendors, top one only 22%) → estimated ~$85K annual savings.
• Forecasting - linear burn-rate projection per category with budget-overrun alerts.
• Receipts, Budgets, Recurring spend, Cross-border FX, Spend profiles, and an AI insights feed round out the /insights, /receipts, and /budgets surfaces.

A Python LangGraph sidecar adds multi-agent reasoning on top of the single-call AI. Four swarms: an approval debate (Prosecutor ‖ Defender → Judge), a fraud investigator (one agent per suspect), a compliance reviewer + false-positive challenger (cuts unnecessary critical phone alerts), and an insights multi-lens sweep (4 lenses → ranker). The sidecar is stateless (the TS routes gather context and persist results + per-agent traces) and degrades gracefully — if it isn't running, every route falls back to the original single-call path. The Agents tab # Brim It - AI Expense Intelligence

Brim Financial × MPC Hacks. An AI-powered expense-intelligence platform for SMB credit-card spending - built on real (anonymized) company-card transaction data and the real Brim expense policy. Make the data talk.

A non-technical finance manager can chat with their company's spend in plain English, manage a digitized expense policy and triage violations, run an AI pre-approval queue, and generate review-ready expense reports - all grounded in real numbers.

Pages: / is a cinematic scroll-reveal brand overview. The app is four surfaces - Overview (Spending · Budgets), Insights, Governance (Violations · Receipts · Audit), and Workflow (Approvals · Reports · Agents) - plus a floating Ask AI chat on every page. The older /dashboard, /compliance, /approvals, /reports, /chat routes redirect into these.

Every report approval, pre-approval decision, and HIGH/CRITICAL compliance alert is notarized on Solana: a SHA-256 of the record's canonical snapshot is written into a Solana Memo transaction (devnet), giving a publicly verifiable Explorer link. A Verify action re-hashes the live record and compares it to the immutable on-chain hash, so any post-approval tampering is provably detectable. Server-side keypair (no wallet needed), env-gated, best-effort (never blocks an approval). See Solana Audit Anchor. Run npm run solana:setup to provision a funded devnet keypair.

Next.js 15 (App Router) · TypeScript · Tailwind + shadcn/Radix · Recharts · better-sqlite3 · Google Gemini (@google/genai) + OpenAI (gpt-4o-mini, via the sidecar) · Brim Agents sidecar (Python · FastAPI · LangGraph · langchain) · Solana (@solana/web3.js + @solana/spl-memo) · ElevenLabs + Twilio (voice) · Spline/three/GSAP (landing) · swr · sonner · zod. Full library list in Submission. Branded in Brim's teal/cyan (#007d93 / #00c1d5).

Critical compliance alerts call your phone via an ElevenLabs Conversational AI agent over Twilio (high/medium/low stay in the in-app bell only). The same agent has no persona - it announces itself as the automated Brim compliance line - and can also answer inbound questions about compliance.

better-sqlite3 needs a persistent filesystem. Deploy to a host with a volume (Render / Railway / Fly) and run npm run db:reset on first boot - see Dockerfile and render.yaml. For a serverless target (Vercel), swap the client in lib/db.ts to Turso/libSQL (SQLite-compatible). Full instructions in Setup & Deploy.