Inspiration

InvestIA was inspired by a simple but critical problem: many people make investment decisions based on intuition, social buzz, or a single asset that they think is attractive, without understanding how much risk they are actually taking. We wanted to build a tool that helps people test their assumptions in a safe, educational, and explainable way. The project was shaped around a very clear idea: decision-making should be transparent, not opaque.

What it does

InvestIA is a simulated investment copilot that helps users evaluate whether a position is compatible with their risk profile. Users can choose a fictional profile, ask questions such as whether a given stock allocation is reasonable, and receive a verdict based on risk exposure, concentration limits, and portfolio constraints. The system does not provide real investment advice or execute transactions. Instead, it shows the maximum acceptable allocation, the rule that breaks first, and the risk contribution of the proposed position.

For example, a user may ask whether it is reasonable to put 30% of a portfolio in Tesla. InvestIA calculates the actual risk burden, compares it to the investor's tolerance, and explains the result using a clear risk framework. The same logic also works for different personas, such as a conservative investor, a more aggressive profile, or an ESG-constrained investor.

How we built it

We built InvestIA as a hybrid system combining deterministic financial calculations with agent-based reasoning and governance checks. The core calculations are implemented in Python and are designed to be transparent and reproducible. These include volatility measures, annualized risk, contribution to portfolio risk, and the point of balance between a user's desired allocation and the maximum safe allocation.

The application layer uses Streamlit for the interface, while AWS services are used in the full production setup. Amazon Bedrock powers the profiling, analytical, governance, and explicability agents. The data layer includes cleaned market data, macro variables, and ESG-related information. Every major decision is logged in a trace so that the final answer can be explained back to the user with evidence.

Challenges we ran into

One of the main challenges was balancing explainability with technical complexity. Financial risk models can become difficult to interpret, especially when many factors such as concentration, volatility, correlation, and ESG constraints interact. We had to ensure that the system did not simply provide a polished answer that sounded convincing, but instead stayed grounded in objective calculations and auditable traces.

Another challenge was making sure the system respected the simulation-only constraint. We intentionally designed strict guardrails to prevent real investment recommendations or transactions. This meant the project had to combine good financial logic with careful governance and validation layers.

We also had to handle real-world data issues, including partial missing information, ESG classification ambiguities, and the fact that the data set is limited to a relatively short historical window. These constraints had to be documented and incorporated into the design so that the tool remains honest about its limits.

Accomplishments that we're proud of

We are proud of building a working system that turns a complicated investment question into a clear, explainable decision. The project demonstrates how an agentic architecture can be used without sacrificing traceability and trust.

We are especially proud of the combination of: a risk-based decision engine, a structured multi-agent workflow, explicit guardrails, and a user-facing explanation layer that links every conclusion back to a trace and a calculation. The fact that a user can ask a question, see the limit of their proposal, and inspect the reasoning is a key strength of the product.

We also successfully created a simulation environment where the product remains educational and non-actionable, which is essential in a finance-oriented application.

What we learned

We learned that in high-stakes domains, the real value is not in producing a persuasive answer, but in producing a defensible and explainable one. The project reinforced the importance of separating calculation, reasoning, and verification.

We also learned that user trust depends on transparency. If a model can explain where a risk limit comes from, which assumptions it used, and which sources it relied on, the answer becomes much more valuable than a simple recommendation. This is particularly important in investment contexts, where the risk of overconfidence is high.

Finally, we learned that constraints matter as much as model performance. Guardrails, data quality checks, and trace logging are not optional extras; they are central to building a credible financial assistant.

What's next for InvestIA

The next steps for InvestIA are to deepen the analytical model and expand the range of supported scenarios. We want to include more macroeconomic factors, broader stress tests, more robust ESG logic, and richer portfolio simulations. We also want to improve the way the tool explains the relationship between a user's profile and the risk contributions of different assets.

Longer-term, we see InvestIA becoming a decision-support platform for exploring portfolio choices in a safe and educational setting. The goal is not to replace human judgment, but to make it more informed, more structured, and more transparent.

Built With

Share this project:

Updates

Submission history