Inspiration

We wanted to help small businesses build and grow further, especially ones that are less tech-savvy and may have disorganized records.

What it does

Small businesses often keep important information across spreadsheets, documents, and records of conversations or past decisions. They may not have the time or money to build a structured database or hire outside advisors, yet they still need to make decisions that affect their growth and finances. Our product lets a business ask a decision question and uses its existing files to help assess the likely long-term effects. It finds relevant information, such as cash flow, workloads, inventory records, and company history; identifies missing information; and explains the potential benefits, costs, and risks, with evidence from the business’s files.

For example, if a business asks, “Should I host a promotional event?”, the product can review the available business data, identify or ask for the assumptions it needs, and compare pessimistic, base-case, and optimistic scenarios. It then summarizes the estimated costs and benefits, reviews risks, and explains what the results depend on.

How we built it

Frontend: HTML, CSS, and JavaScript Backend: Python and FastAPI AI: OpenAI API with GPT-6 Luna Agent Tools: Model Context Protocol (MCP) Database: SQLite and SQLAlchemy Data Processing: pandas, openpyxl, and pypdf Validation: Pydantic Testing: pytest and Node.js

Challenges we ran into

We had challenges with running into token limits and tool usage limits. It was also difficult for us to turn messy business records into reliable inputs. Over the course of the implementation, we had to reduce our scope and start making more assumptions.

Accomplishments that we're proud of

We're happy that we were able to build a working flow from uploaded records to financial comparisons. ClearStep combines AI interpretation with deterministic calculations, source evidence, and user confirmation in a simple interface.

What we learned

We learned that when working with AI, we would balance the tradeoff between determinism and taking advantage of AI's flexibility. During implementation, we realized we need to prioritize clear boundaries, reliable calculations, and visible assumptions in order to make it easier to understand decisions and trust the agent's answer.

What's next for ClearStep

We want to expand the scope and test it with more inputs to improve extraction across different records. Right now, we mostly focused on our demo data, but we would like to scale this up and make it more robust. In addition, we want to support more types of queries in order to support more financial decisions with ClearStep.