“Afford today. Protect tomorrow.”
This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle
What I Built
What if your financial assistant didn't just tell you how much money you have, but helped you decide whether you should spend it right now?
That question inspired me to build KAIROS, an AI-powered financial decision assistant designed to help people make smarter purchasing decisions based on their actual financial circumstances.
Traditional budgeting applications are excellent at tracking expenses, categorizing transactions, and visualizing spending habits. However, they often fail to answer the question that matters most when someone is about to make a purchase:
"Can I afford this today without putting my future financial stability at risk?"
For freelancers, students, gig workers, and people with irregular income, answering that question is especially difficult. Having enough money in your bank account today does not necessarily mean you can safely afford a purchase.
A payment might arrive late. Rent might be due next week. An unexpected expense might consume your emergency savings.
KAIROS approaches personal finance as a decision-making problem rather than simply an expense-tracking problem.
How KAIROS Works
A user describes a purchase they are considering, provides their financial information, and asks KAIROS whether making that purchase is financially sensible.
KAIROS combines AI-powered understanding with mathematical financial analysis to produce actionable recommendations.
The system is designed around several core capabilities:
1. Natural-Language Financial Understanding
Users can describe their financial situations conversationally instead of manually completing complicated financial forms.
For example:
"I want to buy a ₹45,000 laptop. I have ₹70,000 saved, my rent is due next week, and I expect a freelance payment later this month. Should I buy it now?"
The AI layer interprets the request and extracts the information needed for analysis.
2. Financial Evidence Extraction
KAIROS's architecture supports extracting structured financial details from natural-language messages and image-based evidence.
The goal is to reduce manual data entry while preserving the ability to inspect and correct extracted information.
3. 90-Day Cash-Flow Forecasting
Rather than checking only today's available balance, KAIROS models the potential impact of a purchase over a 90-day financial horizon.
It considers information such as:
Current savings and available balance.
Expected income and payment dates.
Recurring financial obligations.
Upcoming expenses.
Emergency reserve requirements.
The proposed purchase amount.
This helps expose future cash shortages that a simple affordability calculation could overlook.
4. Multiple Purchasing Strategies
KAIROS evaluates alternatives instead of treating every financial decision as a simple yes-or-no question.
Its decision engine can compare:
Buy Now: Purchase immediately if the financial constraints permit it.
Partial Purchase: Reduce or restructure the proposed expenditure.
Installments: Evaluate a payment schedule against future cash flow.
Wait: Delay the purchase until a safer financial date.
Adjust Spending: Explore whether changes to discretionary expenditure make the purchase feasible.
5. Deterministic Financial Validation
One of the most important architectural decisions is separating AI interpretation from financial arithmetic.
Language models are useful for understanding ambiguous financial requests and explaining recommendations, but numerical safety checks require consistency.
KAIROS uses deterministic financial calculations to validate candidate plans against the modeled cash-flow constraints.
This creates a more dependable foundation than asking a language model to invent a recommendation solely from a conversational prompt.
6. Explainable Recommendations
Instead of producing a mysterious financial score, KAIROS aims to communicate why a purchase is considered affordable, risky, or better postponed.
The user should understand how upcoming obligations, expected income, and minimum reserve requirements influence the recommendation.
Demo
Code
How I Built It
I built KAIROS using a hybrid architecture that combines open-weight AI models, local inference, a custom Python orchestration pipeline, and deterministic financial computation. The goal was to create an intelligent financial assistant that understands complex financial situations without relying on AI-generated arithmetic for critical decisions.
KAIROS integrates an optional, fully local AI layer powered by Ollama, using two specialized open-weight models:
Qwen3 4B: Interprets natural-language financial queries, understands purchasing intentions, and extracts financial information from conversational inputs.
Qwen2.5-VL 3B: Processes visual evidence, including financial documents, receipts, and images, to identify relevant financial details.
Ollama: Enables local model execution without requiring a proprietary cloud AI API.
Pydantic: Validates model-generated JSON against structured financial data schemas.
Python: Powers the custom orchestration pipeline, financial forecasting, scenario evaluation, and deterministic calculations
Why Does Open Innovation Matter?
Open innovation is fundamental to KAIROS because financial intelligence should be accessible, privacy-conscious, transparent, and independent of proprietary AI providers.
By integrating Ollama, Qwen3 4B, and Qwen2.5-VL 3B, I was able to build an AI-powered financial decision assistant that supports local inference without requiring users to send sensitive financial information to third-party cloud AI services.
Here's what open innovation made possible:
Privacy-First Financial Intelligence: Open-weight models allow KAIROS to process financial conversations, receipts, and documents locally, giving users greater control over sensitive financial data.
Freedom from Vendor Lock-In: Unlike systems built around a single closed API, KAIROS can replace, upgrade, or customize its AI models without rebuilding the financial computation engine.
Lower Operating Costs: Local inference eliminates per-request cloud API charges, making it more practical to develop an affordable financial assistant for students, freelancers, and independent professionals.
A closed API could provide similar language-understanding capabilities, but it would not offer the same level of control over local execution, model customization, deployment independence, and inference costs.
Top comments (0)