F.old: Precision Sourcing Powered by Gemini

An autonomous agent that reads seller descriptions to filter junk, calculates real logistics costs, and finds high-ROI resale deals 24/7.

Comment

Inspiration

The world of "flipping" (buying used electronics to resell) is a race against time. The biggest problem isn't finding items; it's filtering the noise.

I realized I was spending 80% of my time manually analyzing bad deals: listing descriptions that hide defects, "bundles" filled with worthless sports games, or cheap items that are actually 50 miles away (where gas costs eat the entire profit margin).

I wanted to build an "Unfair Advantage." I didn't just want a search engine; I wanted a Financial Analyst. I wanted an agent that thinks like a veteran reseller: cynical about descriptions, precise about logistics, and obsessed with Net Profit. Thus, F.old was born.

What it does

F.old is an autonomous sourcing agent powered by Gemini 3. It acts as a 24/7 "Sniper" for the used electronics market.

Instead of just showing search results, F.old generates a Financial Dossier for every listing:

Semantic Audit: It reads messy, unstructured seller descriptions to detect hidden "red flags" (e.g., "untested" usually means broken) or identify high-value keywords (e.g., distinguishing a rare "Pokémon Crystal" cartridge from a worthless "FIFA 08").

Predictive Logistics: It doesn't just calculate distance; it calculates Cost. It pulls the user's location, the item's location, and current fuel prices to determine the exact cost of pickup.

ROI Calculation: It subtracts the item cost + fuel cost from the estimated market value to show the True Net Profit.

If a deal looks good but requires a 2-hour drive that destroys the margin, F.old marks it as ROI: NEGATIVE. If a listing is messy but contains a rare item, F.old marks it as OPPORTUNITY.

How we built it

We built the core using Python and the Gemini 3 API.

The Brain (Gemini 3): We used Gemini's advanced natural language processing to parse unstructured marketplace data. We engineered prompts that force the model to act as a skepticism engine, analyzing seller intent and categorizing item conditions (New, Used, Junk) based on text nuances.

The Logistics Engine: We integrated geolocation data to calculate real-world travel distances, which are dynamically converted into financial metrics based on average MPG and fuel prices.

The Interface: A terminal-based "Live Feed" that mimics a high-frequency trading desk, giving the user instant "Buy" or "Pass" signals.

Challenges we ran into

The biggest challenge was Contextual Value. To a standard search engine, a "Gameboy with 5 games" is always better than a "Gameboy with 1 game." But to a reseller, that "1 game" might be worth $100 (Pokémon) while the "5 games" might be worth $5 total (Sports titles). Teaching Gemini to recognize specific high-value titles within a messy text block—and ignore the "shovelware"—was difficult. We had to refine our system prompts multiple times to ensure the AI understood that quantity ≠ quality in the retro market.

Accomplishments that we're proud of

The "Junk Filter": We successfully tuned the agent to identify and reject "low-value bundles" that often trick human beginners.

Real Logistics Math: Seeing the agent correctly reject a cheap item because the gas money would make the deal unprofitable was a huge "Eureka" moment. It proved the agent understands the business, not just the product.

Speed: Turning a 45-minute manual search and spreadsheet calculation into a sub-second autonomous decision.

What we learned

We learned that Gemini is incredibly capable at "reading between the lines." It could pick up on subtle seller hesitation in text descriptions (like "worked last time I checked") that usually indicates a broken item. We also learned that logistics is the silent killer of resale profits. By automating that calculation, we fundamentally changed how we look at local inventory.

What's next for F.old: Precision Sourcing Powered by Gemini

Market Expansion: Moving beyond electronics into furniture and vintage clothing.

Auto-Negotiator: Implementing an agent that can draft the perfect opening message to a seller based on the listing's age and sentiment.

The Affiliate Revenue Engine: This is our endgame for monetization. We are building a referral ecosystem where F.old takes a commission on successful flips found through the platform. We project this affiliate model will generate 78.8% of our net profit, turning F.old into a self-sustaining financial platform rather than just a subscription tool.

Official API Partnerships: To scale beyond the prototype phase, our priority is securing official API access from major marketplaces (like eBay, Mercari, and Meta). This moves us from "scraping" to "direct integration," ensuring 100% data stability, legal compliance, and real-time push notifications for new listings.

Built With

Share this project:

Updates