Agents
EV Deal Finder
818 used EV listings ranked by deal quality. Find the best one.
Pulls live used EV listings from eBay Motors and Carvana, normalizes them into a single dataset, scores each listing on 5 weighted factors, and surfaces the best deals in a filterable dashboard with direct links to live listings.
Quick Start
Inputs
- EV models Vehicle makes and models to search on eBay Motors and Carvana (e.g., Tesla, Nissan Leaf, Chevy Bolt, Kia EV6).
Outputs
What you get after a full run.
- 818 used EV listings from eBay Motors and Carvana, scored and ranked
- Deal score (0–1) per listing based on 5 weighted factors
- Price vs. mileage scatter chart, color-coded by deal score
- Deal score distribution by model
- Filterable, sortable listings table with direct links to live eBay and Carvana listings
- CSV export of the full ranked dataset
How it works
A 6-phase pipeline. Read the blog here for a deeper explanation.
- Search eBay SERP agent pulls listing IDs across 10+ EV model queries — Tesla, Nissan Leaf, Chevy Bolt, Kia EV6, Hyundai Ioniq, Rivian, and more. Each search query targets a specific EV model with a ‘used’ condition filter applied at the query level, not in post-processing. The agent returns item IDs, titles, prices, and listing URLs — enough to identify which listings are worth fetching in full.
- ExtractBatch extract fetches all eBay listing pages in parallel and parses Item Specifics — make, model, year, mileage, condition, and title status. Item Specifics is the structured table eBay sellers fill out per listing — it’s where make, model, trim, mileage, condition grade, and title status live. Batch extract fires all requests concurrently, so the full set of listing pages is fetched in roughly the time it takes to fetch one.
- Buildagent Carvana had no pre-built Nimble agent. A new one was created from scratch using Nimble Agent Builder — no CSS selectors, no manual scraping logic. The agent was built by pointing Nimble Agent Builder at a Carvana search results page and defining the output schema visually — price, make, model, year, mileage, condition, and listing URL. The builder handles selector generation, pagination, and render timing automatically.
- CollectCarvana agent paginates 30+ pages of live EV inventory and returns structured listing data per page. Each page call returns up to 20 listings. The agent follows Carvana’s pagination by incrementing the page parameter until no new listings are returned, and all results are written to a single flat list before normalization.
- Normalize818 listings from both sources merged into a unified schema: source, make, model, year, price, mileage, condition, title status, city, and listing URL. Normalization handles the main cross-source inconsistencies: eBay uses full state names while Carvana uses abbreviations; mileage appears as a formatted string on eBay and as an integer on Carvana; condition grading uses entirely different scales. All price fields are cast to float and null-checked before scoring.
- Score5-factor deal score computed per listing — price (30%), mileage (25%), year (20%), condition (15%), title status (10%). All scored relative to the full dataset, not per model. Each factor is scored on a 0–1 scale relative to the full dataset — not per model or per source — so a cheap Honda Clarity competes directly with a cheap Tesla. The final score is a weighted sum; weights were tuned empirically against a manually labeled set of 50 listings that were clearly good or bad deals.
Stack
Nimble primitives plus the full runtime stack.
APIS & AGENTS
What it does
- ebay_search_results_community_2026_03_23 Pulls eBay Motors search results and returns item IDs for individual listings across any EV model query.
- extract_batch Batch-fetches individual eBay listing pages in parallel — extracts Item Specifics including make, model, year, mileage, condition, and title status.
- carvana_electric_cars_serp_2026_04_30_auw1p8rf Custom agent built with Nimble Agent Builder. Paginates Carvana’s EV inventory and returns structured listing data per page.
RUNTIME STACK
Role
- Nimble Agent Builder Used to create the Carvana agent from scratch — schema defined visually, no CSS selectors written.
- streamlit Dashboard with sidebar filters, price and mileage charts, and a ranked listings table.
- plotly Price vs. mileage scatter chart, deal score distribution by model, listings by year bar chart.
- python 3.9+ Scoring engine, data normalization across two sources, and batch extract orchestration.
- MIT license Fork, modify, ship — no restrictions.
Reach out if you have any questions.
Talk to an Expert