Your Personal
AI Investment
Advisor

Overview

WealthMind is a personal AI-powered investment advisor built specifically for Jay and Jo Marie. It replaces the need for a traditional financial advisor by combining real-time market data, portfolio tracking, and conversational AI guidance — delivered primarily through Telegram with a web dashboard for visual reporting.

The app is designed for two people who are new to investing, want to make informed decisions as a couple, and need a system that slows them down before acting on impulse — while being approachable enough that both partners will actually use it daily.

Product Goals

This app exists to solve three specific fears and one operational need.

• Prevent emotional decisions. Markets move. People panic. The agent is the calm voice that puts volatility in context before you do something you'll regret.
• Educate through action. Every recommendation comes with a reason. You learn by doing, not by reading textbooks. Lessons are tied to real decisions with real money.
• Keep both partners aligned. Investing as a couple means agreeing on risk and goals before acting. The investment vote feature forces that conversation.
• Replace the research grind. No more bouncing between Yahoo Finance, Reddit, and news sites. One conversation, real data, AI synthesis.

Success Metrics

In 90 days, success means: Jay and Jo Marie are making sound decisions based on data and conversation from the platform — and the portfolio is growing.

Behavioral Metrics (Sound Decisions)

• Agent consulted before every trade. Every investment decision is logged with a confidence score. Did you ask before you acted?
• Investment vote completed. Both Jay and Jo Marie logged a vote before any move was made. Alignment rate tracked over time.
• No panic sells. Zero reactive decisions during market downturns without the agent being consulted first.
• Weekly engagement. Both users actively using the agent at least 3x per week. Jo Marie's usage is the real benchmark.

Financial Metrics (Profit)

• Portfolio performance vs S&P 500. Primary benchmark. Matching or beating over 12 months is a strong year-one result.
• Starting balance tracked. $1,000 starting point logged. Every contribution and withdrawal recorded for accurate return calculation.
• Goal progress. At least one financial goal defined and showing measurable progress by 90 days.

Users

Jay

Operator and builder. Sets up the system, defines the financial profile, and uses the agent for deeper market research and portfolio analysis. Comfortable with data and complexity. Wants directness, real recommendations, and no hedging.

Jo Marie

Mobile-first. Accesses the agent via Telegram on her phone. Needs plain language, personality, and an agent that feels like a trusted friend — not a financial textbook. Jo Marie is the usability benchmark. If she's not using it, the product has failed regardless of how technically sound it is.

Features — V1

Conversational AI Agent (Telegram)

• Natural language conversation. Ask anything — "Should we buy Apple?" or "What happened to the market today?" — and get a real, reasoned response in plain English.
• Financial context awareness. The agent knows your goals, holdings, risk tolerance, and timeline before answering anything. Every response is personalized.
• Push-back capability. If you're about to make a decision that doesn't fit your profile, the agent tells you and explains why. It will tell Jay no.
• Emotion check mode. When markets drop and panic sets in, the agent responds with data and calm perspective before any action is taken.
• Learning layer. Short explanations tied to real decisions you're making — not textbook theory. You learn because of what's happening to your actual money.

Market Research (Perplexity Finance)

• Real-time stock data. Prices, charts, key metrics retrieved on demand within conversation.
• Earnings summaries. Live earnings call transcripts summarized in plain English within minutes of release.
• SEC filing search. Ask about a company's 10-K or S-1 in natural language.
• "Should I buy this?" tool. Paste a ticker. The agent evaluates it against your profile and gives a clear recommendation with reasoning.
• News synthesis. Market-moving news filtered for relevance to your portfolio and summarized without jargon.

Web Dashboard (Visual Layer)

• Portfolio snapshot. Current holdings, total value, performance over time. Updated from Postgres as agent conversations happen.
• Goal tracker. Named goals with target amounts, timelines, and visual progress bars. Numbers tied to something real.
• Contribution planner. "If we add $X per month, here's where we'll be in Y years." Simple math, powerful motivation.
• Research archive. Every "Should I buy this?" report saved and accessible. Full history of agent recommendations.
• Investment vote tracker. Log showing every decision, both votes, and the outcome. Accountability over time.
• Weekly summary view. Monday morning portfolio health check — what changed, what it means, what (if anything) to consider.

Couple Features

• Shared portfolio view. Both Jay and Jo Marie see the same dashboard and conversation history.
• Investment vote. Before any move, both partners log a vote via Telegram. Agent waits for alignment before proceeding.
• Couple-aware agent. The agent knows both users individually and factors both perspectives into guidance.

AI Agent Design

The agent is the core of the product. It runs on Claude API with its own system prompt, financial persona, and tool access. It lives completely outside of Cerebro and JARVIS — this is a personal agent, not a business agent.

Agent Identity

Agent Persona

The agent behaves like a seasoned, candid financial advisor who knows Jay and Jo Marie personally. It speaks plainly, gives real recommendations, explains its reasoning, and pushes back when needed. It does not hedge everything or pepper responses with disclaimers. It has personality — it's allowed to be funny.

Agent Tools

• Market data lookup. Calls Perplexity Finance API for real-time stock prices, financials, and news on demand.
• Portfolio context retrieval. Reads financial profile and current holdings from Postgres before every response.
• Goal evaluation. Scores any investment decision against active goals and risk profile.
• Dashboard writer. Posts summaries, research reports, and portfolio updates to Postgres for the web dashboard to display.
• Weekly summary generation. Triggered via n8n every Monday morning. Sent via Telegram and posted to dashboard.

When the Agent Is Wrong

The agent will make bad calls. Markets are unpredictable and no advisor — human or AI — bats 1.000. When a recommendation doesn't work out, the agent should acknowledge it plainly, explain what the data said at the time, and update its reasoning going forward. A feedback flag in Telegram lets Jay or Jo Marie mark any response as unhelpful, which queues a system prompt review.

Multi-Agent Research Team

The main agent doesn't do all the research itself. Behind every conversation is a full team of specialist agents — each one running the best AI model for its specific domain. Jo Marie never sees this. She just gets one smart, synthesized answer.

The Team

Orchestration Flow

Fallback Logic

When a specialist model is unavailable, the system degrades gracefully — not catastrophically. Each specialist has a defined fallback so Jo Marie never hits a dead end.

The Goldman Sachs Parallel

How It Works at Goldman Sachs or Morgan Stanley

A senior advisor at a top firm doesn't do all the research themselves. They have an entire team behind them — and the client only ever talks to one person.

The client only ever talks to one person — the senior advisor. That person synthesizes everything the team produces into a clear recommendation. That's exactly what you're building. Jo Marie talks to the main agent. Behind that conversation is a full research team — each specialist doing what they do best, routing results back to the orchestrator, who synthesizes it into one clean answer.

The Cost Difference

The One Honest Gap

Human advisors have something the AI doesn't have yet: relationships. A senior advisor knows when a client is scared even if they don't say it. They read tone, body language, and history built over years of in-person meetings.

The main agent approximates this through conversation history and pattern recognition — and will get better at it over time. But the gap is worth acknowledging. It's not a reason not to build. It's a reason to keep improving the system prompt as you learn how Jay and Jo Marie actually communicate.

Jo Marie Onboarding

The first interaction is everything. If Jo Marie's first message gets a generic, robotic response — she disengages permanently. The agent's first message is scripted to be warm, funny, and immediately personal.

Onboarding Flow

• First message. Agent introduces itself with personality. Immediately establishes it's on Jo Marie's side.
• Profile setup conversation. Agent walks Jo Marie through her goals, risk tolerance, and timeline conversationally — not a form. Takes 5-10 minutes.
• First recommendation. Based on her profile, the agent immediately gives one concrete piece of guidance. Shows value before she's even fully set up.
• Dashboard introduction. Agent sends the web dashboard link and explains what she'll find there.

Financial Profile

The financial profile is what separates this from a generic chatbot. Stored in Postgres and injected into every agent conversation. Without it, the agent is just a market search tool. With it, every response is personalized to your actual situation.

Notification Strategy

The agent should proactively reach out at the right moments — not so often it becomes noise, not so rarely it feels absent.

Scheduled Notifications

• Monday morning summary. Weekly portfolio health check at 8:00 AM. What changed, what it means, what (if anything) to consider this week.
• Monthly goal progress check-in. "You're X% toward your [goal]. At this pace, you'll hit it in Y months." Keeps goals visible.

Event-Triggered Notifications

• Significant portfolio drop. Any holding drops more than 10% in a day. Agent reaches out proactively with context before panic sets in.
• Earnings reports. For any company you hold, agent sends a plain-English summary within 30 minutes of earnings release.
• Major market event. S&P 500 moves more than 2% in either direction. Agent contextualizes it relative to your portfolio.
• Investment vote pending. If Jay votes and Jo Marie hasn't yet — gentle reminder after 24 hours.

Architecture

Dual-interface system. Telegram handles conversation and notifications. The web dashboard handles visual reporting. Both are driven by the same agent and read from the same Postgres database.

Tech Stack

Security Model

Two-phase security approach. V1 is simple and effective for personal use. V2 adds production-grade authentication when the web dashboard matures.

Hosting Strategy

Development — Mac Mini M4

Build and test locally. Docker ensures the local environment matches production exactly. No surprises when deploying.

Production — Railway

Cloud deployment for mobile access. Jo Marie opens the dashboard in Safari on her iPhone like any website. Telegram bot runs as a separate Railway service. Both read from the same Postgres instance.

Cost Model

PM Gaps — Open Items

Items identified that need decisions before or during the build. Tracked here until resolved.

Build Phases

V2 Roadmap

Features intentionally excluded from V1 to keep the build focused and realistic for a small team.

• Fidelity API live sync. Replace manual CSV with real-time portfolio data. Automatic updates without Jay touching anything.
• Auth0/Clerk authentication. Magic link + MFA + passkeys for web dashboard. Already planned — just not needed until V1 is stable.
• Native mobile app. Wrap the dashboard in a React Native shell for App Store distribution. Only needed if expanding beyond Jay and Jo Marie.
• Crypto integration. Add crypto market data once traditional investing fundamentals are solid. Don't mix asset classes too early.
• Tax optimization. Track cost basis, flag tax-loss harvesting opportunities. Requires CPA review before any tax guidance goes live.
• Push notifications (web). Browser push in addition to Telegram alerts. Lower priority — Telegram already handles this well.
• Datacenter migration. Move from Railway to Center Square for cost reduction and full infrastructure control.