Trading-card identification & pricing engineer

We’re building a consumer app for collectors: scan a card (sports + TCG), get an instant ID and current value. Visual matching already works well — our last-mile problem is data quality. The gap: Turning a visual match into clean, canonical card data — year, brand, set, player, card #, parallel/variant, grader, grade. Today names and years get mangled and variants get dropped, which then breaks exact price matching. What we want help with: • A robust scan → canonical-card pipeline — entity resolution against a real   card catalog/reference, not just string-parsing listing titles. • Tighter comp selection so pricing keys off the right card (sold comps   first, active listings as fallback). Stack: Python/Django + Celery, Postgres/Redis, Next.js/TypeScript, eBay API, Gemini/LLM, plus a few card-pricing data sources. Ideal fit: • Familiarity with the hobby is a must — parallels, short prints, graded   slabs, year-on-card vs. season, and the messy edge cases. • Familiarity with other card-scanning apps is a bonus. • Built card recognition or data-normalization / entity-resolution systems   before is a plus.