Artificial Intelligence Engineer (Project-based)

Company Description: JMarketing is a premier creative advertising agency with a strong presence across North America and the Middle East. The team delivers bespoke solutions in branding, design, digital marketing, and web design & development for a diverse portfolio of clients. By combining creativity with data-driven strategy, JMarketing helps organizations strengthen their market position and reach targeted audiences effectively. The agency values innovation, collaboration, and high-quality execution in every campaign, offering professionals the opportunity to work on impactful and international projects.
Project Description: JMKTG is building a new AI photo experience and microsite for a client activation in the Gulf, and we're looking for a freelance developer to build it with us.
The concept in short: a browser-based camera experience where a user takes a selfie, an AI model returns a branded transformed portrait, and the result comes back to them as a downloadable, shareable asset. We're designing and art directing the whole thing in-house. We need someone to build it properly.
What you'd be building
The front end from scratch, bilingual Arabic and English, built RTL-first so neither language is an afterthought.
The generation pipeline behind it: job queue, polling, retries, graceful failure, and branded in-experience error states.
Integration with image generation APIs, including the character and face consistency work that makes the output usable rather than generic.
Deployment on real cloud infrastructure with a staging environment and documented handover, sized for a busy activation night rather than a demo.
Full funnel instrumentation so the client can see completions, drop-off by step and language split while the activation is running.
A consent and data handling layer built correctly for facial data under Saudi law.
What you need
Core engineering
Strong JavaScript and TypeScript. Comfortable in vanilla JS and in at least one modern framework (React/Next.js, Svelte/SvelteKit, Vue/Nuxt or Astro).
Browser media and graphics: getUserMedia and the MediaDevices API, ImageCapture, Canvas 2D and OffscreenCanvas, Blob and object URL handling, the Web Share API, and the File System Access API.
Node.js on the back end (Express, Fastify or NestJS) or Python with FastAPI. Go is fine too.
Real-time progress delivery using WebSockets or server-sent events, not naive polling.
Job queues and async orchestration: BullMQ with Redis, SQS, Inngest, Temporal or Celery. Idempotency keys, retries with backoff, circuit breakers and backpressure.
Postgres or MongoDB, Redis, and object storage with signed URLs and lifecycle rules (S3, R2 or GCS).
PWA and service worker experience, with offline and degraded-network resilience. Venue wifi is the enemy.
Generative AI
Production integration with image generation and editing APIs. Gemini image models, OpenAI GPT-Image, Flux, Imagen, Stability, Ideogram, Leonardo or Recraft.
Serverless model hosting and custom pipelines on Replicate, fal.ai, Modal, RunPod or Hugging Face Inference.
Identity and character consistency is the make-or-break skill here: InstantID, IP-Adapter, PuLID, InsightFace, or LoRA and DreamBooth fine-tuning for a locked brand look.
ControlNet for pose, depth and edge conditioning. Inpainting, outpainting and instruction-based editing.
ComfyUI workflow authoring and custom node graphs, or equivalent reproducible pipeline tooling.
Seed control, deterministic reruns, prompt versioning and A/B testing of generation quality.
Face detection and landmarking with MediaPipe, InsightFace or face-api.js. Segmentation and matting with SAM or rembg.
Restoration and upscaling: Real-ESRGAN, GFPGAN or CodeFormer.
Content safety: NSFW and minor detection, moderation APIs, and hard guardrails. This is a family brand, and the output cannot embarrass anyone.
Bilingual and RTL
Arabic-first bidirectional builds. CSS logical properties, correct dir switching, bidi isolation, and no LTR assumptions baked into layout.
i18n tooling: i18next, FormatJS or next-intl, with externalized strings rather than hardcoded text.
Arabic typography judgement: appropriate type families, line height, and Arabic-Indic versus Western numeral handling.
Accessibility to WCAG 2.2 AA, including keyboard navigation, ARIA and screen reader behaviour in both languages.
Infrastructure and reliability
Cloud deployment on AWS, GCP, Cloudflare, Vercel, Fly.io or Railway. Docker, and Kubernetes or ECS if the load profile calls for it.
Infrastructure as code with Terraform or Pulumi, CI/CD through GitHub Actions, and clean separation of staging and production.
CDN and image delivery: Cloudflare, CloudFront, Cloudinary or imgix. Modern formats and sensible caching.
Autoscaling and capacity planning, with load testing using k6 or Artillery before we go live rather than after.
Observability: Sentry, Datadog or Grafana; uptime monitoring; and alerting that actually reaches someone during an activation.
Privacy and security
Consent implemented properly: a real gate before the camera opens, granular opt-ins, and consent state logged and auditable rather than assumed.
Retention and deletion that actually execute. Automated lifecycle rules, verifiable deletion, and no orphaned copies sitting in storage, logs, caches or the CDN.
Working awareness of Saudi PDPL and SDAIA, particularly that facial imagery counts as sensitive personal data and that data residency requirements may constrain which generation APIs we can use.
Security ownership of what you build: OWASP Top 10, CSP, HSTS, input validation, secrets management, signed URLs, and no personal data in URLs or query strings.
Willingness to go through a multinational client's vendor security review and remediate what it surfaces.
Measurement
GA4 and Google Tag Manager, including server-side tagging and Consent Mode v2.
Product analytics with PostHog, Mixpanel or Amplitude, built around a real funnel rather than pageviews.
Custom event schema design and a dashboard the client can open mid-activation without calling us.
Strong pluses
Image-to-video and motion generation: Runway, Luma, Kling, Veo, AnimateDiff or Stable Video Diffusion.
On-device inference in the browser with TensorFlow.js, ONNX Runtime Web or WebAssembly, to cut round trips.
WebGL, WebGPU or three.js for richer in-browser treatment.
CRM and messaging integration: Salesforce Marketing Cloud, Braze, HubSpot, Twilio, or the WhatsApp Business API, which carries real weight in this market.
The engagement
Freelance and project-based, with an ongoing maintenance retainer available for the right person. Remote, but you need genuine overlap with Arabia Standard Time (UTC+3) and a Sunday-to-Thursday working week.
Start is immediate once the quote is agreed and signed.
Send your portfolio plus two examples of bilingual AR/EN work, if applicable, to hello@thejmarketing.com

Work arrangement
Yes