Numa Nova

Numa Nova transforms financial reconciliation with AI-powered analysis and voice conversations, cutting resolution time by 70%.

Comment

NumaNova: AI-Powered Voice Reconciliation Assistant

Inspiration

Financial reconciliation remains one of the most tedious, time-consuming processes in accounting, with finance teams spending up to 30% of their month-end close time hunting down and resolving transaction discrepancies. During a stint in corporate finance, I witnessed firsthand how accountants would spend days sending emails and making calls to resolve simple mismatches. The revelation came when our CFO mentioned that a 5-minute call often resolved what 20 emails couldn't. This inspired us to reimagine reconciliation as a voice-first, AI-powered experience that could eliminate the back-and-forth and dramatically accelerate financial operations.

What it does

NumaNova transforms financial reconciliation through a unique combination of AI analysis and voice communication:

  1. Intelligent Discrepancy Detection: The system automatically identifies unmatched transactions and categorizes discrepancies
  2. AI-Powered Analysis: Using Perplexity's research capabilities, FinVoice analyzes patterns to determine likely causes (timing differences, missing documentation, etc.)
  3. Autonomous Decision-Making: The AI decides which stakeholders to contact based on discrepancy type and urgency
  4. Proactive Voice Outreach: Rather than generating another email, FinVoice initiates voice conversations via Vapi to quickly resolve issues
  5. Comprehensive Documentation: All resolutions are documented for audit trail and future pattern recognition

How we built it

We built FinVoice as a tightly integrated system leveraging multiple cutting-edge technologies:

  • Next.js & TypeScript: Created a responsive, enterprise-grade frontend interface
  • Claude AI & Perplexity MCP: Implemented the reasoning layer that analyzes transaction patterns and determines causes
  • Vapi MCP: Enabled natural voice conversations for reconciliation resolution
  • Claude Desktop: Served as the orchestration layer connecting our AI components -Gumloop: Process PDF and match reconciliation between invoices and shipping receipts -Rime: As tutorial and welcome on the landing page -Vizcom: Hero image on our website

The technical architecture revolves around a central MCP integration hub that manages communication between different specialized AI systems. Transaction data flows from accounting systems into our analysis engine, which then triggers appropriate voice-based workflows based on discrepancy type and severity.

Challenges we ran into

Building FinVoice within the hackathon timeframe presented several significant challenges:

  1. Integrating Multiple MCP Servers: Coordinating between Perplexity and Vapi MCP servers required careful design of the central communication hub
  2. Creating Natural Voice Flows: Designing voice conversations that balance accounting precision with natural dialogue proved complex
  3. Simulating Real Financial Data: Generating realistic test scenarios that covered various reconciliation edge cases
  4. Handling Decision Complexity: Implementing the decision logic for when to initiate calls versus other forms of communication
  5. Time Constraints: Building a full-featured demo within the hackathon timeframe required focused prioritization

Accomplishments that we're proud of

Despite the challenges, we achieved several significant milestones:

  1. Successfully built an end-to-end reconciliation system in under 4 hours
  2. Created a seamless integration between multiple MCP servers
  3. Developed realistic mock data that demonstrates common reconciliation scenarios
  4. Implemented an autonomous AI decision-making flow for resolution prioritization
  5. Designed a professional UI that finance teams would actually want to use

What we learned

This project provided valuable insights into the future of AI agents and voice technology:

  1. The power of specialized AI systems working together through MCP
  2. The efficiency of voice communication for resolving complex financial discrepancies
  3. The importance of domain-specific knowledge in financial AI applications
  4. The challenges of building multi-agent systems with different capabilities
  5. The potential for voice-first interfaces in traditionally text-heavy domains

What's next for NumaNova

We see tremendous potential to expand NumaNova's capabilities:

  1. ERP Integrations: Direct connections to QuickBooks, Xero, NetSuite, and SAP
  2. Multi-Currency Support: Handling complex FX reconciliation scenarios
  3. Learning Capabilities: Improving accuracy through continuous learning from resolved cases
  4. Expanded Voice Capabilities: Adding more specialized personas for different reconciliation scenarios
  5. Mobile Integration: Allowing finance teams to manage reconciliation on the go

Built With

Share this project:

Updates