Brianna

Inspiration

Small and medium-sized businesses often lack the financial oversight tools available to large enterprises. Managers are expected to monitor spending, enforce expense policies, approve purchases, and understand budget health—often while juggling many other responsibilities.

We wanted to explore how AI could act as a finance analyst for SMBs: helping managers understand spending patterns, identify compliance risks, make approval decisions faster, and generate meaningful reports from real financial data.

That vision became Brianna.

What it does

Brianna is an AI-powered expense intelligence platform that helps managers monitor company spending, enforce policy compliance, approve purchase requests, and generate reports from natural-language conversations.

Built around six months of real anonymized trucking-company transaction data, Brianna combines analytics, compliance monitoring, and AI reasoning into a single dashboard.

Natural Language Analytics

Managers can ask questions such as "Which categories are driving spending this month?" or "Compare fuel expenses between employees." Brianna translates natural language into SQL, analyzes the resulting data, and automatically selects the most appropriate visualization to present insights.

Policy Compliance Engine

Brianna continuously evaluates transactions against company expense policies and identifies potential violations. The system detects missing pre-authorizations, repeat offenders, split-charge attempts, policy threshold violations, and high-risk spending behavior. Violations include severity scoring, policy citations, and contextual reasoning.

AI-Powered Approval Workflow

Employees submit purchase requests through a simple form. Managers receive employee spending history, budget impact analysis, AI-generated recommendations, and context-aware reasoning—allowing them to approve or deny requests in a single click.

Automated Reporting

Brianna generates both executive budget reports and employee spending profiles. Reports include narrative summaries, spending breakdowns, policy findings, approval outcomes, and budget health assessments.

Budget Tracking

An animated fill bar shows total spend vs budget, color-coded by utilization threshold, alongside a category pie chart with human-readable labels (Fuel, Meals, Vehicle Maintenance). Updates every 5 seconds as approvals come in. A year-end forecast projects monthly burn rate forward, showing projected total, surplus or overrun, and a two-tone progress bar (actuals solid, projected translucent), color-coded based on whether year-end spend lands under 85%, under 100%, or over budget.

Employee Ranking

Brianna evaluates and ranks all employees via a compliance leaderboard using a scored model that combines pre-auth violations, split-charge detections, denied requests, and spend vs team average. This helps managers identify top performers and those requiring closer oversight.

How we built it

The frontend is built with React, Vite, TypeScript, Recharts, and Tailwind CSS, providing dedicated manager and employee experiences with real-time updates and interactive visualizations.

The backend uses Node.js, Express, and TypeScript, exposing analytics, compliance, approvals, reporting, and chat APIs while coordinating all AI workflows.

For AI, we used Claude (Anthropic) with multi-step reasoning chains, structured JSON generation, and Zod validation. We implemented a four-stage AI workflow: intent extraction, SQL generation, contextual analysis, and visualization selection. We also use AI for approval recommendations, compliance reasoning, employee insights, and reporting.

The data layer uses a SQLite in-memory database loaded with 4,000+ anonymized transaction records via an Excel ingestion pipeline, enabling safe read-only SQL analysis in real time.

Challenges we ran into

One of the biggest challenges was safely allowing AI to query financial data. Since Claude generates SQL dynamically, we enforced strict read-only constraints using validation layers before executing any query.

Another challenge was ensuring structured outputs for charts, reports, and workflows. We solved this using Zod schemas and retry mechanisms.

We also built more advanced compliance logic that considers employee history, spending behavior, and context rather than simple rules. Finally, balancing AI flexibility with system reliability required careful design of validation and error-handling pipelines.

Accomplishments that we're proud of

Built a conversational analytics system powered by real financial data. Designed a multi-step AI reasoning pipeline. Implemented contextual compliance scoring. Built a full AI-assisted approval workflow. Added anomaly and fraud detection including split-charge detection. Created an employee ranking and insights system. Unified everything into a single dashboard. Generated automated financial reports. Delivered a polished decision-focused UX.

What we learned

Building reliable AI systems requires more than calling an LLM. We learned how to design structured reasoning pipelines, validate outputs, integrate AI with databases, and balance flexibility with reliability.

We also learned that visualization and workflow design are just as important as the AI itself. The most valuable systems help users make decisions, not just generate text. Most importantly, we learned how to use AI as a decision-support tool rather than a chatbot.

What's next for Brianna

We see Brianna evolving into a complete AI financial operations platform. Future plans include corporate card integrations, Slack and Microsoft Teams workflows, receipt OCR and matching, automated low-risk approvals, advanced forecasting models, department-level planning tools, vendor optimization insights, multi-manager enterprise support, real-time spending alerts, and continuous compliance monitoring.