Building an AI-Powered Personalized Financial Education Platform
An AI-powered financial education platform for the Philippine market. Personalized onboarding, AI-generated reports, and interactive that makes financial clarity accessible to everyday users.
Industry: Consumer Finance / AI / FinEdTech
Scope: Product Strategy, UX Design, AI Agent Architecture, Full-Stack Development, Analytics
Stack: Next.js · MongoDB · LangChain · Claude (Anthropic API) · Xendit · PostHog · Vercel
Market: Philippines (B2C)
Overview
Perry is an AI-powered financial education platform designed for the Philippine market. It helps everyday users better understand their financial situation through a personalized onboarding flow, AI-generated financial insights, and a chat-based coaching experience.
Perry is a financial education and clarity tool. It does not provide financial advice, recommend specific financial products, or act as a licensed advisor.
The Opportunity
Most financial tools available to Filipino consumers are either generic content (blogs, YouTube, budgeting templates) or professional services priced for people who already have wealth to manage. For someone earning ₱20,000 to ₱40,000 a month, neither option works.
The gap isn't information. It's personalization. People don't need another article about budgeting rules. They need a tool that can take their actual numbers and help them see their situation more clearly.
Perry was built to fill that gap: a financial companion that's accessible, personalized, and available 24/7 through a chat interface.
The Product Experience
Perry's user journey starts with a guided onboarding flow designed to feel like a conversation rather than a form. Users walk through a series of screens that collect financial context in plain, approachable language. The tone is warm and non-judgmental, which matters when users are sharing sensitive financial details.
That context feeds into Perry's AI agent system, which generates personalized financial insights based on the user's actual data. Users receive structured, readable output that reflects their specific situation, not generic content.
From there, users can continue interacting with Perry through a chat interface. Perry retains context across sessions, handles follow-up questions, recalculates outputs when the user's situation changes, and maintains a consistent, supportive tone in conversational Filipino-English.
All outputs are educational in nature. Perry helps users understand their finances better. It does not make product recommendations or provide regulated financial advice.
AI Agent Architecture
This is where the core technical work happened. Perry's AI layer is not a simple prompt-and-response wrapper. It's a multi-agent system built with LangChain, designed to handle the complexity of turning unstructured user input into structured, personalized outputs.
Why Agents, Not Just Prompts
A single prompt call can generate text, but it can't reason across multiple domains, retrieve context from different sources, decide what tools to use, or structure its output differently depending on the user's situation. Perry needed all of that.
A user with ₱45,000 monthly income and ₱12,000 in loan payments needs a very different output than someone earning ₱25,000 with no debt but no savings. The AI couldn't just fill in a template. It had to assess the situation, decide which aspects to focus on, and structure the output accordingly.
How the Agent System Works
The system is built as a set of specialized agents orchestrated through LangChain:
Context Assembly Agent. Pulls together the user's full financial profile from onboarding data and previous interactions stored in MongoDB. Structures the raw input into a normalized context object that downstream agents can reason over.
Financial Analysis Agent. Takes the assembled context and performs the core reasoning: identifying patterns, calculating key ratios, and determining which areas to surface. Operates within strict guardrails to keep outputs educational and never cross into regulated advice territory.
Output Structuring Agent. Takes the analysis and formats it into the appropriate output type. Each output type has its own structure and tone requirements. This agent handles the difference between a first-time comprehensive output and a quick follow-up response.
Conversation Agent. Manages the ongoing chat experience. Maintains conversation history, handles follow-up questions, decides when to trigger a full re-analysis versus responding from existing context, and manages tone calibration across sessions.
Guardrail Layer. A cross-cutting system that reviews all agent outputs before they reach the user. Enforces boundaries: no specific product recommendations, no investment advice, no claims of expertise. Also ensures outputs stay relevant to Philippine financial context and use appropriate language.
For follow-up questions, the Conversation Agent can short-circuit this flow when full re-analysis isn't needed, pulling from cached context and prior outputs for faster responses.
Why This Architecture Matters
This approach meant Perry could handle a wide range of user situations without brittle conditional logic or hundreds of prompt templates. The agents reason, adapt, and structure their outputs based on the actual data in front of them. The system is also modular: improving the analysis logic doesn't require rewriting output formatting, and tightening the guardrails doesn't break the conversation flow.
Technical Stack
Next.js (App Router) for the frontend. Server components, API routes, and a responsive interface optimized for mobile, which was critical for the Philippine market.
MongoDB as the primary database. User profiles, financial data, conversation history, and agent context. The document-based model was a natural fit for the varied structure of financial profiles where each user's data shape differs based on their situation.
LangChain for AI agent orchestration. The multi-agent system was built on LangChain's agent framework, using tool use, memory management, and chain composition. Claude Opus as the underlying language model.
Xendit for payment processing. GCash, GrabPay, Maya, bank transfers, and card payments. Mobile wallet support was essential for the Philippine market.
PostHog for product analytics. Event tracking across the full user journey, with feature flags and session replay for experimentation and debugging.
Vercel for hosting and deployment.
Results
Perry acquired thousands of users and was able to scale rapidly, serving a demographic that traditional financial services largely overlooks: young to mid-career Filipino workers navigating financial complexity without access to professional guidance.
- Thousands of personalized financial outputs generated, each shaped by the user's actual data.
- Positive user reception, with users consistently reporting that Perry helped them see their financial situation more clearly.
What This Project Demonstrates
Perry required product strategy, UX design, AI agent architecture, payment infrastructure, analytics, and iteration based on real usage. Not a prototype or demo, but a product that processed real transactions and delivered value to real users.
It demonstrates StackSpace's ability to:
- Design and build multi-agent AI systems using LangChain
- Build AI products grounded in real user needs, not just technical capability
- Translate AI outputs into experiences that feel clear, trustworthy, and useful
- Integrate local payment infrastructure for Southeast Asian markets
- Ship production AI products using agentic development workflows with Claude
Have a similar challenge?
Book a blueprint session and we'll map your workflows, scope the build, and tell you what it'll take.