Thirteen Labs

Like Dave Ramsey in your pocket. Talk to your money, understand your spending, never miss a bill, and reach your goals in plain English.

Comment

Thirteen Labs

Like a Dave Ramsey in your pocket. Talk to your money, understand your spending, never miss a bill, and reach your goals in plain English.

Inspiration

From passive tracking to active understanding.

You check your bank app. Numbers stare back. What do they mean? What should you change?

Personal finance tools show people what they spent, but not what it means or what to change. Money management is a job in itself.

Most apps rely on manual tracking and static dashboards that users must interpret themselves. In reality, people don't need more charts they need clear guidance tied to their actual behaviour. That mental load is the real friction.

We wanted to build a financial companion that understands spending automatically and explains it in simple terms. By using real transaction data as the source of truth and combining it with AI reasoning, we aimed to shift personal finance from passive tracking to active understanding.

Financial clarity should be conversational and continuous not something users have to calculate on their own. Thirteen Labs flips the model: real transaction data + AI reasoning = guidance you can talk to.

What It Does

Thirteen Labs is an AI personal finance assistant that works with your Solana wallet. You connect your wallet and the app syncs your transactions. Those transactions are categorized into three buckets Now (day-to-day spending), Commit (bills and subscriptions), and Grow (savings and investments) and shown in a hero pie chart so you see how your actual spending lines up with your plan.

Instead of digging through dashboards, you ask questions. A chat bar sits at the bottom of the app. This financial data is analyzed by this intent based multiple AI system with a unified memory layer that identifies patterns, detects unnecessary expenses, and generates personalized guidance from all the sources and planning you do on our app. The assistant doesn't just answer questions it explains your spending, your bills, and your goals in plain language, with numbers preserved. Ask what matters to you; get answers that direct action.

One more issue we tackle is missing your bill payments. There is a periodic reminder system that keeps you on track and reminds you to pay your credit card bills!

You can interact conversationally with the assistant (text or voice) to ask questions like:

  • Where am I overspending?
  • What expenses can I reduce?
  • How can I save faster?
  • Am I on track for my goals long term?
  • Which bill should I pay first?
  • Am I at risk of missing any payments?

Voice interaction: Speak to the assistant (mic → STT → chat → TTS) and hear replies — hands-free guidance while cooking, commuting, or multitasking, with no voice files stored.

Financial Horizons Model

We structure finances across three horizons each with a clear benefit for you:

Now — day-to-day discretionary spending (dining, groceries, outings)
See where your day-to-day money goes so you can trim what doesn't matter.

Commit — recurring obligations (rent, subscriptions, bills)
Never miss a credit card bill or subscription. The assistant surfaces what's due and when, so you stay on top of obligations instead of paying late fees or hurting your credit.

Grow — long-term savings and investments
Track how much actually reaches savings and investments not just intentions.

The hero chart shows how your actual spending maps to these three horizons so you see the gap between plan and reality at a glance. One look tells you whether you're living within your means or overspending before goals get funded.

How We Built It

Thirteen Labs uses transaction ingestion plus an intent-routed orchestrator and Backboard (Gemini).

Flow:

Transaction Source (Solana) → Categorization (SPL memo + fallback) → Financial State (DB snapshot) → Intent-Routed Orchestrator → Backboard (Gemini) → Assistant Output

The Solana wallet acts as a real-time transaction feed that automatically generates financial history without manual entry. Transactions are categorized via SPL memo parsing into investments, bill_payments, and short_term_goals, with fallback heuristics for raw activity.

The orchestrator builds a condensed financial context (transactions, bills, goals, budget allocation) and routes user intents:

  • Spending / Transactions — recent categories, amounts, specific cuts
  • Bills / Obligations — unpaid bills, due dates, payment order
  • Goals / Progress — named goals, progress %, deadlines
  • General insights — data-driven summary and actionable steps

Backboard provides the LLM with memory and conversation continuity. The assistant is constrained to personal finance and produces TLDR explanations with preserved numeric values. We extensively used backboard in our project wherever we needed to integrate with an assistant.

Challenges We Ran Into

Converting raw transactions into meaningful financial categories
Transaction data is low-level and inconsistent. We designed normalization and categorization logic to map raw activity into human-understandable spending behavior. We learned that SPL memo conventions plus fallback heuristics are essential when source data varies.

Designing safe, finance-scoped AI outputs
Financial guidance requires guardrails. We constrained the assistant to personal finance explanations and numeric fidelity via system prompts and modes (conservative/balanced/aggressive) to avoid unsafe or speculative advice.

Multi-agent orchestration consistency
Our pipeline uses intent routing to specialize behavior across goals, bills, transactions, and insights. Maintaining consistent context across these paths required structured financial state passing in a single orchestrator.

Explaining behavior instead of just analyzing it
We focused on interpretable insights rather than metrics. This required designing TLDR output constraints and layman-language translation in prompts.

Vibe coding hell One of the unexpected challenges we encountered was vibe-coding feedback loop hell. We believe AI has significantly raised the ceiling of what can be built within 24 hours, the limiting factor is no longer implementation capability. Like many teams, we found that the real difficulty lay in deciding what not to build. The rapid pace of ideation, prototyping, and iteration created a cycle where new possibilities constantly surfaced, making it easy to overextend scope or pivot prematurely. This required deliberate restraint, prioritisation, and repeated refocusing on core objectives to maintain progress toward a coherent deliverable rather than an ever-expanding feature set.

We learned that the hard part isn't fetching data it's turning it into advice users can act on without a finance degree.

Accomplishments That We're Proud Of

Real-time syncing with Solana Wallet Connect your wallet once, and your financial history builds itself. Transactions are automatically ingested, categorized, and reflected instantly inside the app, transforming raw blockchain activity into structured financial insight.

Voice interaction with no storage
Speak to the assistant and hear replies hands-free guidance while cooking, commuting, or multitasking, via ElevenLabs STT/TTS, with all audio in-memory and no voice files stored.

Multi-agent financial planner (Backboard + Gemini) Using Backboard’s persistent memory layer and Gemini’s reasoning capabilities, our system routes intent across specialized agents that reason over a shared financial state. Agents communicate, reconcile constraints, and refine recommendations before responding, delivering coherent, finance-scoped guidance instead of isolated chatbot answers.

Financial understanding without manual input
Connect a transaction source and instantly get behavioral insights no manual tracking, no data entry. Your spending history builds itself.

Never miss a bill
The bill payment horizon surfaces what's due and when, so you avoid late fees, interest spikes, and credit score hits from forgotten credit card or subscription payments.

Finance-scoped conversational assistant
An AI assistant that explains your spending, your savings, and your allocation in simple TLDR terms grounded in real numbers — advice tailored to you, not generic tips.

Single orchestrator with intent routing
One API call, context-aware responses for goals, bills, transactions, and general advice no separate microservices, just smart routing and structured context.

Multi-horizon financial model (Now / Commit / Grow)
We created a framework that evaluates short-term spending, recurring obligations, and long-term wealth simultaneously.

What We Learned

Interpretation is more valuable than visibility.
Clear explanations create more value than dashboards.

Financial guidance requires strict AI constraints.
Finance-scoped prompts and numeric fidelity rules improved reliability.

Behavior change depends on contextual insights.
Users respond better to goal-oriented explanations than raw totals.

Real transaction data improves AI relevance.
Insights grounded in actual behavior are more actionable and trustworthy.

What's Next for Thirteen Labs

Our vision: a financial intelligence layer that sits alongside everyday activity and continuously guides decisions toward user goals no extra work required.

We plan to expand Thirteen Labs into a continuous personal finance copilot:

  • multi-account transaction aggregation
  • connect and make payments through Solana from our app.
  • Add more personality to agents
  • dynamic chart creation and updation using natural language
  • biometric authentication
  • Integrate maps to give local experiences suggestions based on spending budget
  • Connecting fiat money providers for example (Paypal)

Built With

  • React JS
  • Python Flask
  • Backboard.io (Gemini LLM enabled multi-agent orchestration pipeline)
  • ElevenLabs (Text-to-Speech and Speech-to-Text for voice interaction)
  • Valkey
  • Solana wallet integration (Auth + transaction ingestion)
  • Vultr (Deployed our AI application)

Tagline

Understand your spending. Reach your goals.

Built With

Share this project:

Updates