Fix camera scanning speed & reliability in existing Lovable web app (live card scanner)

I have a fully working web app called FLAMEZ — built in Lovable (React/TypeScript) — that lets collectors scan trading cards live via their phone camera, get them auto-identified, and track their collection. The core app, database, pricing, and social/recording features all work well. I need an experienced developer to improve the camera and recognition experience on both front and rear cameras, with front camera set as the default. Use case context: the app is used two ways — solo scanning via rear camera (highest quality/accuracy), and a social "with friends" mode via front camera, propped up phone-style, so people can film themselves/friends reacting while opening packs, cards held up to the front camera to scan. Front camera should be the default active camera on launch, but both need to work excellently — this isn't "primary vs afterthought," both are core to the product. What's already working (please don't rebuild): Live camera capture with card recognition via the Ximilar API (already integrated and authenticated) Confidence-based accept/reject logic for scan results, with separate sharpness thresholds already tuned per camera Session recording and auto-clip generation, works identically regardless of active camera What I need improved: Front camera as the default on app launch, with an easy toggle to switch to rear camera. Recognition speed — currently averaging ~0.7-1.2 seconds per card on rear camera; want both cameras performing at a similarly fast, consistent standard. Looking for genuine improvements: image compression/payload tuning, capture pipeline efficiency — not just UI tricks. Recognition consistency on both cameras — needs real diagnosis (lighting, focus distance, capture timing, resolution constraints) rather than just loosening confidence thresholds. Front camera in particular needs realistic tuning given typical selfie-camera hardware limits (lower resolution, fixed focus). Camera/recording quality — both live feed and recorded footage should be as sharp and reliable as possible on each camera type. Ideal freelancer: Strong experience with the Lovable platform specifically (or comfortable working directly in a React/TypeScript/Vite codebase it produces) Real experience with browser camera APIs (getUserMedia, MediaRecorder), including front/selfie camera focus behavior and constraints — not just general web dev Comfortable diagnosing an existing system rather than only writing new code from scratch Please don't apply if your plan is to just loosen the recognition confidence threshold without addressing actual image capture quality — I've already been through that dead end and need someone who understands the difference. What I'll provide: Full access to the working Lovable project Existing Ximilar API access A written summary of what's already been tried and what's working/not working, so you're not starting blind Timeline: asap