👻 Phantom

We don't Monte Carlo market returns. We Monte Carlo you.

An AI-powered behavioral financial planning system that predicts how your financial habits—not just the market—affect your future wealth.

Traditional financial planners assume you'll stick to the plan.

Phantom assumes you won't.

Instead of asking "What if the market crashes?", Phantom asks:

• What if you skip your SIP after three months?
• What if lifestyle inflation catches up?
• What if impulse purchases become more frequent?
• What if your financial discipline slowly drifts over time?

Using a multi-agent AI architecture, Phantom analyzes spending behavior, builds a personalized financial roadmap, simulates hundreds of possible behavioral futures, and continuously adapts as your habits evolve.

---

Why Phantom?

Financial planning today focuses almost entirely on market uncertainty.

Human behavior causes far more financial plans to fail than market volatility.

Phantom shifts the focus from predicting markets to predicting people.

Traditional Financial Planning	Phantom
Assumes perfect discipline	Models real human behavior
Static financial plan	Adaptive behavioral planning
Market-based simulations	Behavior-based simulations
One-time recommendations	Continuous monthly monitoring
Generic advice	Personalized AI-generated guardrails

---

Demo

Frontend: (Add deployment link here)

Backend API: (Add deployment link here)

Video Demo: (Add YouTube link here)

---

Features

Behavioral Fingerprinting

Extracts a 14-dimensional Behavioral Signature Vector (BSV) from user transactions.

Examples include:

• Spending consistency
• Present bias
• Impulse tendency
• Savings discipline
• Goal persistence
• Lifestyle inflation risk
• Financial resilience

---

Financial Archetype Detection

Classifies every user into behavioral archetypes such as:

• Optimist Procrastinator
• Guilt Investor
• Lifestyle Maximizer
• Goal Chaser
• Stability Seeker

Each archetype includes:

• Confidence score
• Strengths
• Weaknesses
• Personalized narrative
• Actionable insights

---

AI Financial Plan

Generates a personalized financial roadmap including:

• Monthly allocation strategy
• SIP recommendations
• Emergency fund planning
• FIRE projection
• Tax optimization
• Milestones

---

Behavioral Monte Carlo Simulation

Unlike traditional Monte Carlo simulations that randomize market returns, Phantom randomizes human decisions.

Examples:

• Skipping investments
• Overspending
• Salary growth
• Unexpected expenses
• Lifestyle inflation
• Motivation decay

500 behavioral futures are simulated to estimate the probability of long-term financial success.

---

Personalized Guardrails

Instead of simply saying "Spend less," Phantom creates commitment systems such as:

• Automatic savings rules
• Spending caps
• Emergency fund triggers
• Investment automation
• Monthly accountability checkpoints

---

Monthly Drift Monitoring

Financial behavior changes over time.

Each month Phantom:

• Re-analyzes transactions
• Detects behavioral drift
• Measures divergence from the original plan
• Updates recommendations
• Suggests new guardrails

---

Example Workflow

User uploads transactions
            │
            ▼
Behavioral Fingerprinting
            │
            ▼
Financial Archetype Detection
            │
            ▼
AI Financial Planning
            │
            ▼
500 Behavioral Simulations
            │
            ▼
Behavioral Guardrails
            │
            ▼
Monthly Drift Monitoring

---

Example

Input

Income:
₹75,000

Goal:
Retire by 45

Behavior:
"I save consistently for three months,
then spend heavily on travel."

↓

Behavioral Fingerprint

Savings Discipline: 61

Impulse Risk: 82

Present Bias: 76

Consistency: 54

↓

Archetype

Optimist Procrastinator
Confidence: 82%

↓

Simulation

500 behavioral futures

Success Probability:
43%

Most likely failure:
Month 7

↓

Guardrail

Automatically transfer bonuses
into emergency savings before
they reach the spending account.

---

Architecture

                     ┌─────────────────────────────────┐
                     │      FastAPI (main.py)          │
                     │ POST /analyze   POST /drift     │
                     └────────────┬────────────────────┘
                                  │
                     ┌────────────▼────────────────────┐
                     │         Orchestrator            │
                     │ Stateful multi-agent pipeline   │
                     └──┬──────┬──────┬──────┬─────────┘
                        │      │      │      │
                        ▼      ▼      ▼      ▼

                 Agent 1  Behavioral Fingerprint
                 Agent 2  Archetype Detection
                 Agent 3  Financial Plan
                 Agent 4  Behavioral Simulation
                 Agent 5  Guardrails
                 Agent 6  Drift Monitoring

---

Multi-Agent System

Agent 1 — Behavioral Fingerprint

Produces a 14-dimensional Behavioral Signature Vector from spending history.

Model

Claude Opus 4.5

---

Agent 2 — Archetype Classifier

Assigns one of eight behavioral financial personalities.

Model

Claude Haiku 4.5

---

Agent 3 — Financial Planner

Builds personalized investment and savings strategies.

Includes:

• FIRE roadmap
• SIP allocation
• Tax optimization
• Milestones

Model

Claude Sonnet 4.5

---

Agent 4 — Behavioral Simulation

Runs 500 behavioral futures to estimate plan success.

Model

Claude Opus 4.5

---

Agent 5 — Guardrail Architect

Designs commitment systems that reduce behavioral risk.

Model

Claude Sonnet 4.5

---

Agent 6 — Drift Monitor

Runs monthly behavioral reassessments.

Model

Claude Haiku 4.5

---

Tech Stack

Backend

• FastAPI
• Python
• Pydantic
• Anthropic Claude API

Frontend

• HTML
• CSS
• JavaScript

AI

• Multi-Agent Orchestration
• Behavioral Reasoning
• Monte Carlo Simulation

---

Repository Structure

phantom-plan-backend/
│
├── main.py
├── orchestrator.py
├── models.py
├── requirements.txt
├── .env.example
│
└── agents/
    ├── agent1_fingerprint.py
    ├── agent2_archetype.py
    ├── agent3_plan.py
    ├── agent4_simulation.py
    ├── agent5_guardrails.py
    └── agent6_drift.py

---

Installation

Clone

git clone <repository-url>

cd phantom-plan-backend

---

Install Dependencies

pip install -r requirements.txt

---

Configure Environment

cp .env.example .env

Add:

ANTHROPIC_API_KEY=YOUR_KEY

---

Run

python main.py

or

uvicorn main:app --reload

---

API

POST /analyze

Runs the complete five-agent behavioral analysis.

Returns:

• Behavioral fingerprint
• Archetype
• Financial plan
• Monte Carlo simulation
• Personalized guardrails

---

POST /drift

Monthly behavioral re-analysis.

Returns:

• Drift score
• Behavioral changes
• Updated recommendations
• New guardrails

---

GET /health

Returns service status.

---

GET /archetypes

Returns all supported behavioral archetypes.

---

Model Routing

Agent	Model	Reason
Behavioral Fingerprint	Claude Opus 4.5	Deep reasoning
Archetype	Claude Haiku 4.5	Fast classification
Planning	Claude Sonnet 4.5	Structured reasoning
Simulation	Claude Opus 4.5	Complex behavioral modeling
Guardrails	Claude Sonnet 4.5	Creative reasoning
Drift	Claude Haiku 4.5	Cost-efficient recurring inference

---

Design Decisions

Why Multi-Agent?

Separating responsibilities makes each component easier to evaluate, improve, and replace independently.

Why Behavioral Simulation?

Most financial plans fail because of inconsistent human behavior rather than poor investment products.

Why Claude?

Different models are selected based on reasoning complexity and cost efficiency.

Why Graceful Degradation?

Every agent has a fallback, ensuring the pipeline continues even if an individual model fails.

---

Current Limitations

• Uses manually provided transaction data
• No direct banking integrations
• No persistent user history
• Monthly monitoring is API-driven
• Prototype optimized for hackathon deployment

---

Roadmap

Integrations

• Setu Account Aggregator
• CAMS
• KFintech

AI

• Fine-tuned open-source behavioral models
• Persistent agent memory
• LangGraph orchestration

User Experience

• WhatsApp reminders
• Push notifications
• Mobile application

Infrastructure

• PostgreSQL
• Authentication
• User dashboard
• Long-term behavioral analytics

---

Contributing

Contributions, feature requests, and discussions are welcome.

Please open an issue before submitting large changes.

---

License

MIT License.

---

Phantom doesn't predict markets.

It predicts the person making the decisions.