Money Mentor AI

Money Mentor AI: your stress-free path to financial freedom - one confident step at a time.

Comment

Inspiration

Imagine you’re drowning.... not in water, but in credit cards, car loans, and payday traps. You’re not alone: over 40% of adults worldwide carry this silent weight. But no one talks about it. Shame keeps us quiet. Fear keeps us stuck. And the advice online? Generic, overwhelming, one-size-fits-none. Money Mentor AI changes that. We’re not another finance app. We’re your personal debt coach - private, judgment-free, and built just for you. You tell us your debts, income, and lifestyle. We scan everything and build a custom escape plan: avalanche, snowball, or a hybrid that fits your life. Along the way, we teach you, in plain English, the terms, the traps, the smart moves. So you don’t just get out of debt… you stay out. Money Mentor AI: your stress-free path to financial freedom - one confident step at a time.

What It Does

Money Mentor.ai is your personal financial coach - calm, private, and ruthlessly practical. It reads your real numbers, designs a plan that fits your life, and walks with you until the last balance hits zero.

Core Capabilities

1) See your full picture - no guesswork.

  • Securely connects to Plaid to pull balances, transactions, loans, and recurring payments.
  • Parses PDF statements to capture missing details like APRs, loan interest rates, and mortgage fees.
  • Cleans and standardizes everything into precise debt profiles you can act on.

2) Build a payoff plan that actually works.

  • Runs three proven strategies-

    • Snowball: smallest balances first (fast wins, strong momentum)
    • Avalanche: highest APR first (minimize interest)
    • Consolidation: one lower-rate payment (simplify + reduce cost)

  • Under the hood, we model every month with: $$B_{t+1}=B_t-\big(P-r\times B_t\big)$$ where \(B_t\) = balance, \(P\) = payment, \(r\) = monthly interest. Clear math. Clear timeline.

3) See it, tweak it, own it.

  • Interactive dashboards in React + Framer Motion - smooth, instant, distraction-free.
  • Drag payment sliders and watch payoff dates, interest totals, and cash-flow charts update live.
  • One click to a detailed PDF report: timelines, total interest, monthly playbook, and personalized insights.

4) Learn the language of money - fast.

  • Plain-English explainers for APR, minimums, utilization, debt-to-income, and more.
  • Gemini AI voice narration turns dry terms into quick, memorable micro-lessons.

5) Ask anything. Get answers that stick.

  • A dual-agent AI Tutor + Coach that explains the “why,” highlights trade-offs, and keeps advice grounded in your actual data; no copy-paste tips, no repetition, no fluff.

Bottom line: Money Mentor.ai doesn’t just tell you what to do. It shows you why it works, lets you shape the plan, and stays with you - one confident payment at a time.

How We Built It

Engineered for clarity, speed, and explainability. Every piece serves one goal: turn messy money data into an action plan you can trust.

Tech Stack

  • Frontend: React + Vite for fast dev loops, Tailwind for utility-first styling
  • Backend: FastAPI (Python) for a snappy, zero-friction API you can deploy anywhere.
  • AI Stack: Gemini AI for reasoning + narration, Google ADK (Agent Development Kit) for reliable A2A orchestration.
  • PDF Parsing: ReportLab for pixel-perfect exports.
  • Finance Integration: Plaid API to securely ingest accounts, balances, loans, and recurring payments.

API Endpoints

Method Endpoint Description
GET /health Backend health check.
POST /api/plaid/create-link-token Create Plaid Link token.
POST /api/plaid/exchange-token Exchange public_token; fetch data.
POST /api/docs/extract Extract APR/min/due from PDFs (ADK).
POST /api/plan/generate Generate repayment plan (ADK).
POST /api/recommender/analyze AI recommendations + reasoning (ADK).
POST /api/report/generate Build comparative report (ADK).
GET /api/tutor/next-lesson Next lesson content (ADK).
GET /api/tutor/quiz Get lesson quiz (ADK).
POST /api/tutor/check-answer Check quiz answer (ADK).
GET /api/tutor/progress Get tutor progress (ADK).
POST /api/tutor/progress/update Update tutor progress (ADK).
GET /api/voice/voices List ElevenLabs voices.
POST /api/voice/generate Generate TTS narration.
POST /api/docs/generate-from-plaid Make synthetic PDFs from Plaid.
POST /api/strategies/calculate Compute payoff strategies.
GET /api/adk/health ADK health.
GET /api/adk/agents/status ADK agents status.
GET /api/adk/workflows/{workflow_id} Workflow status.
GET /api/files/{filename} Serve uploaded/generated files.
POST /upload-pdf Upload a PDF.
POST /api/parseur/extract Legacy mock extraction.
POST /api/plaid/connect Link Plaid accounts.
GET /api/plaid/accounts Retrieve accounts/balances.
GET /api/reports/export Download PDF report.
POST /api/insights/voice Voice narration via Gemini.

Multi-Agent ADK Pipeline

Agent Role
Data Agent Fetches, validates, and normalizes Plaid + PDF into standardized debt profiles.
Computation Agent Runs payoff math, interest accrual, and timeline simulations across strategies.
Chat Agent Explains trade-offs and surfaces insights grounded in user data.
Voice Agent Converts key takeaways into clear, natural narration (TTS).
Visualization Agent Streams results to charts, sliders, and dashboards for instant feedback.
DocumentExtractionAgent Parses PDFs for APR, balance, minimum due, and loan fields.
VerifierAgent Sanity-checks and normalizes extracted fields and dates.
PlanGeneratorAgent Builds Snowball, Avalanche, and Consolidation plans.
RecommenderAgent Picks the best plan and explains why.
ReportGeneratorAgent Drafts the comparative, tabular report.
TutorAgent Delivers lessons, quizzes, answer checking, and progress tracking.

All agents communicate over the A2A message bus (ADK).

A2A flow (at a glance): Ingest → Normalize → Simulate Strategies→ Explain → Visualize → Export. The agents coordinate through Google’s ADK A2A fabric for low-latency handoffs and consistent context - so every screen, sentence, and sound bite stays in sync with your numbers.

Challenges We Ran Into - and How We Turned Them Into Momentum

1) Missing vitals behind Plaid.

Plaid shows balances and names but hides the real vitals-APR, minimums, and due dates-inside messy PDFs with promo footnotes. Without those, payoff math drifts and confidence erodes) Solution: Parse statements and verify APR/min/due before planning.

2) CORS walls.

The browser blocked cross-origin calls when origins and preflights didn’t align, and WebSocket requests failed by default. This broke the app-API handshake) Solution: Use a strict allowlist and a proxy that supports HTTP and WebSockets.

3) Unstable WebSockets.

Streams dropped during peak moments and agents stalled, while duplicate events risked double state updates. Reliability and order were shaky) Solution: Use wss, and exponential backoff.

4) Inconsistent PDFs.

Banks format fields differently and totals sometimes disagree, letting OCR noise pollute numbers. Extraction quality varied run to run) Solution: Layered parsers with regex fallbacks and sanity checks gated by confidence.

5) Chatbot drift and repeats.

The chatbot repeated itself, forgot what it said, and slipped into generic advice. It drifted away from the user’s data) Solution: Ground on the latest snapshot with history awareness and style guardrails.

6) Interest math mismatches.

Compounding and rounding rules varied by issuer, and promos or deferred interest skewed totals. Numbers didn’t match statements) Solution: Standardize compounding/rounding, and validate the numbers using heuristics.

7) Charts vs. sliders jitter.

Rapid input desynced axes and tooltips, and animations ran before state settled. Visuals contradicted the data) Solution: Single timeline source, and tick snapping after state settles.

The Outcome

Fewer surprises. Cleaner data. Charts that obey the math. A voice that teaches, not teases. And a plan you can trust because every step - extraction, simulation, explanation - starts with the truth and carries it all the way through.

Accomplishments That We’re Proud Of

  • Fixed Missing APR Data: Integrated Parseur PDF parsing alongside Plaid to fill missing fields.
  • Bug-Free A2A Pipeline: Achieved a fully synchronized multi-agent system with no deadlocks or race conditions.
  • Smart Chatbot Replies: Implemented contextual memory to prevent repetitive answers.
  • Secure Infrastructure: Used proper CORS middleware and HTTPS configurations for safe API calls.
  • Mathematically Sound Strategies: Validated that each repayment strategy followed real financial logic and formulas.
  • Fluent Voice Narration: Tuned Gemini AI parameters for clear and concise financial speech output.
  • Accurate Dynamic Plots: Fixed mapping functions and ensured real-time updates with user sliders.
  • Reliable Report Exports: Ensured async data consistency so every report reflects real-time values.

What We Learned

  • The importance of team communication in defining MVP scope - focusing on working prototypes before scaling features.
  • Practical understanding of Google ADK’s multi-agent orchestration and asynchronous communication.
  • Deep hands-on experience with Plaid API, FastAPI, and Parseur for data integration.
  • Learned how to design AI prompts for explainability and accuracy in financial domains.
  • Improved frontend collaboration using React, and Tailwind.
  • Gained experience balancing performance, explainability, and security in fintech systems.

What’s Next for Money Mentor AI

  • Investment & Savings Advisor: Expand beyond debt management into personalized investment guidance.
  • Reinforcement Learning Agent: Learn from user behavior to recommend optimal monthly payments.
  • Multilingual Support: Add voice and text translations for global reach.
  • Mobile Companion App: Bring Money Mentor AI to iOS and Android.
  • Enhanced Security: OAuth-based Plaid token rotation and encrypted local caching.
  • More Data Sources: Integrate additional APIs (e.g., Yodlee, QuickBooks) for broader coverage.
  • Habit Tracking: Track monthly goals and reward progress toward debt-free milestones.
Share this project:

Updates