VaultFlow

Inspiration

Honestly, I was looking at marketplaces like MicroAcquire and Flippa and realized just how much time is wasted staring at mind-numbing P&L spreadsheets. Private Equity firms have armies of $200k/year analysts to hunt for distressed digital assets, calculate LTV:CAC ratios, and model out buyouts. Everyday entrepreneurs don't.

I wanted to democratize that power. The "aha" moment was realizing that with the right autonomous architecture, I could completely automate the "first pass" of M&A due diligence. I didn't want to build just another dashboard; I wanted to build an AI business partner that actually tells you what to buy and helps you execute the deal.

What it does

VaultFlow is an autonomous Micro-PE Analyst. You drag and drop a messy financial ledger (like a CSV of Shopify or Amazon FBA stores) into the app.

Behind the scenes, VaultFlow maps the data, writes its own SQL queries, and mathematically isolates the "Golden Goose" : a business with great unit economics but terrible operational bloat (e.g., high revenue, but bleeding cash on bad ads).

Instead of making you read a massive report, it streams a conversational audio "voice note" pitching you the acquisition strategy. Finally, it dynamically provisions three professional .xyz domains and links you straight to the checkout to launch the Special Purpose Vehicle (SPV) to buy the company. Complex data in, acquired businesses out.

How I built it

The frontend is React 19 and Tailwind CSS, gated by Clerk. But the real magic is my 100% serverless AWS backend.

I couldn't just wire an LLM to a persistent database (hello, prompt injection and idle compute costs). Instead, when you upload a file, my API Gateway tosses the payload into an SQS queue. A Lambda Worker wakes up, pulls the CSV from S3, and mounts it into a highly ephemeral, in-memory SQLite database (:memory:).

I use Amazon Nova Lite to act as the "Quant Analyst" writing the SQL, and Amazon Nova Sonic to generate the low-latency audio briefing. Once the math is done and the .xyz domains are secured, the sandbox incinerates itself, and the frontend pulls the final Plotly charts and audio from DynamoDB.

Challenges I ran into

Where do I start?

1. API Timeouts: AWS API Gateway has a hard 29-second timeout limit. You can't profile a 50,000-row CSV, run an LLM, and generate audio in 29 seconds consistently. I had to rip out the synchronous architecture at the last minute and replace it with an async SQS-polling loop.
2. LLM Hallucinations: LLMs are notoriously bad at writing perfect SQL on the first try. They invent columns or mess up JOINs.
3. Domain Wrangling: Getting the LLM to strictly output .xyz domains (and not hallucinate .ai or .com) required me to build a custom Python regex failsafe to ensure the sponsor integration was absolutely bulletproof for the UI.

Accomplishments that I'm proud of

I am incredibly proud of the Critic Loop. Because LLMs write bad SQL, I built a self-healing engine. If my in-memory SQLite throws a Python traceback error, the Lambda catches it, feeds that exact error back to Nova Lite, and forces the model to rewrite its own broken code. It loops this until the query executes cleanly. The user never sees the retry process. They just get the right answer.

I'm also really proud of the design. Shifting away from a generic "SaaS" look to a dark-mode, cyan-accented "Bloomberg Terminal" aesthetic made the whole project feel like institutional-grade software.

What I learned

I learned that executives and founders don't want to read dashboards, they want to listen. Decoupling the visual data on the screen from the conversational audio script completely changed how the app feels. It taught me that multimodal outputs (Charts + Voice) are the future of Business Intelligence. I also learned the hard way that serverless AI requires letting go of synchronous API calls early in the development process.

What's next for VaultFlow

Right now, VaultFlow relies on CSV uploads. The immediate next step is building live OAuth integrations with Stripe, Shopify, and Plaid so the platform can continuously monitor deal-flow in real-time. I also want to integrate automated legal template generation, so when you buy that .xyz domain, VaultFlow automatically spins up the Letter of Intent (LOI) to send to the seller.

Built With

• amazon-bedrock
• amazon-dynamodb
• amazon-web-services
• react
• vite
• xyz