Backend engineer

Computer Vision & Backend Engineer (60-Day Build)Company: Wow Nom Type: Fixed-term contract (60 days, full-time) — extension possible Location: Remote (Singapore Time, APAC-friendly hours)How to applyEmail with subject “60-Day CV & Backend Build — Wow Nom” and include:A shipped CV project (repo/demo) + one latency and one accuracy number you achieved and howAvailability to start within 1–2 weeks and timezone(Optional) A brief note on grams estimation from depth vs. monocular on plated dishesMission (60 days)Deliver a production-ready photo recognition system that powers a calorie-counting app end-to-end:Upload → Analyze → Nutrition: From a food photo, return { name, grams, confidence, tags, ingredients, macros } per item, with meal totals and remaining daily targets.Retraining option: Design and ship the infrastructure that learns from user corrections (renames, grams/macros edits) and can retrain/evaluate safely.What you will build (end-to-end scope)Public APIsPOST /api/vision/upload (multipart JPEG/PNG/Web P) → { name, grams, confidence, tags }()POST /api/coach/photo → persist image, call vision, run lookup Food, return items, meal totals, remaining Daily, and coach ReplyFood analysis (multi-cuisine)Gate + Instances: YOLOv8/11 detect (food vs distractors) → YOLO-seg (retina masks)Naming: Sig LIP/CLIP (or compact Vi T) on mask crops, synonyms/taxonomy awareSafety: OOD detector + low-confidence suggestions; safe abstain (no hallucinations)Portioning (grams)Device-depth first (if present), monocular fallback (Mi Da S/Zoe Depth), tabletop plane-fit, coverage %, density lookup (Redis), portion_source=device|mono|heuristicNutrition & ingredientsMap labels → canonical taxonomy (≤400 dishes)Query our nutrition DB or external sources (e.g., FDC) to assemble ingredients + per-ingredient macros, scale by grams, compute meal totalsRetraining loop (feedback → model)Capture user edits & low-margin/OOD crops → store to Click House/S3Scripts & jobs to rebuild datasets, fine-tune, evaluate with metric gates, and publish new artifacts safelyOps & safetyCI evaluator (Top-1/Top-5, OOD FP rate, Portion MAPE, latency SLOs) that blocks regressionsObservability: structured logs, per-stage ms, model/taxonomy versionsPrivacy: consent gate, retention/“delete my images” flow60-Day milestone plan (acceptance-driven)Week 1–2 (Foundation & API)Stand up GPU Fast API /infer-v2 + Node /api/coach/photoReturn stubbed payload matching contract; basic telemetry; dockerizedDemo: curl upload → JSON schema exactly matches app contractWeek 3–4 (Models & Portions)YOLO gate+seg (export ONNX); CLIP/Sig LIP naming with temperature scalingDepth-aware grams (device depth) + mono fallback; density via RedisDemo: multi-cuisine sample set returns names + grams within sanity boundsWeek 5 (Nutrition & Safety)Taxonomy (≤400) + nutrition mapping (our DB / FDC)OOD abstain with suggestions; ingredients + per-ingredient macros scaled by gramsDemo: App-ready payload { name, grams, confidence, tags, ingredients, macros } per item; meal totals & remaining DailyWeek 6–8 (Retraining + CI gates + Canary)Feedback capture from user edits; dataset rebuild scripts; fine-tune pathEvaluator + CI gates (json report) and shadow/canary rollout togglesPrivacy & retention wired; runbook + handover docsFinal Demo (Day 60): end-to-end flow on staging GPU; retrain on a small corrected set; CI passes; canary toggle readySuccess metrics (set at kickoff; used by CI gate)Quality: Top-1 on core ≥ target; OOD FP ≤ target; Portion MAPE ≤ target on depth imagesLatency: p50 ≤ 350 ms, p95 ≤ 800 ms on our staging GPUReliability: CI gate prevents regressions; logs/metrics complete; consent & retention enforcedMinimum qualificationsShipped computer-vision systems to production (beyond notebooks)YOLO detect/seg training or fine-tuning; export to ONNX/Tensor RT and debug opsets/dynamic shapesCLIP/Sig LIP or Vi T classifier work (fine-tune + temperature scaling); OOD thresholdingDepth pipelines (device + monocular), geometric reasoning (plane fitting, coverage)Production APIs (Fast API/Node), Redis/Click House (or similar), Docker, Git Hub ActionsObs/ops: structured logging, latency profiling, privacy/retention patternsNice-to-havesTriton Inference Server, FAISS/ANN, K8s/Helm, W&B/MLflowNutrition data integration (FDC or equivalent), taxonomy designTech you’ll touchPy Torch, Ultralytics YOLOv8/11, SAM/SAM2, Sig LIP/CLIP, Mi Da S/Zoe Depth, ONNX Runtime (CUDA EP), Tensor RT (nice), Fast API, Node/Express, Redis, Click House, Docker, Git Hub Actions.What we provideGPU access (cloud, H100/A10/T4), seed datasets & taxonomy draft, staging infra, and rapid product feedbackClear API contract and benchmark packs for CI gatingHow to applyEmail with subject “60-Day CV & Backend Build — Wow Nom” and include:A shipped CV project (repo/demo) + one latency and one accuracy number you achieved and howAvailability to start within 1–2 weeks and timezone(Optional) A brief note on grams estimation from depth vs. monocular on plated dishes