Platform Architect - AI

Job Description

We're looking for an experienced Platform Architect (AI) to design, implement, and scale AI-driven solutions across our organization. This role combines deep technical expertise with strategic thinking, translating business needs into concrete, production-ready AI architectures. The work spans several domains and technology stacks - computer vision, language models, document processing, and data-driven automation. Each project will bring its own constraints, its own trade-offs, and often its own toolset.

The first project in scope is a visual similarity search/embeddings/vector retrieval solution for automated verification of store display compliance.

Responsibilities:

Design end-to-end architecture for AI/ML systems, from proof of concept to production

Select appropriate technologies, frameworks, and models

Oversee integration of AI solutions with existing systems and data infrastructure

Evaluate build vs. buy decisions for AI capabilities, including cost modelling and vendor lock-in risk

Ensure solutions are scalable, secure, cost-efficient, and compliant with relevant regulations (e.g. GDPR, EU AI Act)

Guide MLOps/LLMOps practices - retraining triggers, drift and quality monitoring criteria, versioning strategy across models and data, and rollback paths and work with MLOps engineers on their implementation

Mentor AI Engineers on AI architecture patterns and emerging best practices

Stay current with the AI landscape (foundation models, vector databases, agentic frameworks, RAG, fine-tuning) and assess their applicability to CK challenges

Create and maintain technical documentation, architecture diagrams, and decision records

Act as the technical liaison to the organization's AI Security function — bringing them in at key design decisions (data flows, model endpoints, third-party integrations) rather than making security calls unilaterally; ensure access control, encryption, and safeguards against adversarial or malicious inputs (e.g. spoofed images, injection attacks on NL components) meet organizational standards, and verify new third-party integrations against existing security approvals rather than assuming blanket coverage from prior vendor vetting.

Project-specific requirements (visual similarity search):

Design and maintain the pipeline architecture: product detection → parallel analysis (visual embedding, OCR, category classification) → hybrid query to the vector database → matching against the planogram

Decide on specific models and components (e.g. RF-DETR vs. alternatives, DINOv2 as embedder), weighing trade-offs: accuracy, cost, licensing, time to delivery

Design category-specific query strategies (e.g. different signal weights for cigarettes vs. bottles - text-signal dominance vs. visual-embedding dominance)

Make hosting and cost decisions: self-hosted GPU infrastructure (AWS) vs. managed services (Roboflow Inference, Pinecone SaaS), appropriate to the scale

Define and oversee the API contract between the AI pipeline and the rest of the system (backend, RELEX integration) - what exactly the model returns, how low-confidence matches are handled

Set measurable quality thresholds (precision/recall per product category, acceptable false-positive rate) jointly with the Product Owner

Verify license compliance of chosen components (e.g. Apache vs. AGPL-3.0 for YOLO-family models) against the product's business model

Provide architectural oversight of the AI/ML Engineer's work - architectural code review, not day-to-day implementation

Plan the improvement loop: how approved image crops from field verification feed back into the index as additional real-world reference material

Engage the organization's AI Security team early to review the pipeline's data flows (store photos, embeddings, RELEX and Pinecone connections) and confirm required controls; verify the RELEX integration follows existing organizational security standards for this specific access pattern (e.g. real-time API vs. batch export), and work with AI Security to define requirements for new components introduced by this project - encryption in transit and at rest, access control for the Pinecone connection, and a data-retention policy for photos that may incidentally capture employees or customers (GDPR relevance given the Polish/EU context).

Requirements:

5+ years of experience in software architecture, with at least 2–3 years focused on AI/ML systems

Strong understanding of machine learning fundamentals, deep learning, and modern NLP/LLM architectures

Hands-on experience with cloud platforms (AWS, Azure, or GCP) and their AI/ML services

Experience with MLOps tools (e.g., MLflow, Kubeflow, SageMaker) and CI/CD for ML pipelines

Proficiency in Python and familiarity with ML frameworks (PyTorch, TensorFlow, Hugging Face)

Experience with vector databases, RAG pipelines, and agentic AI frameworks

Solid grasp of data architecture, APIs, and system integration patterns

Excellent communication skills - able to explain complex technical concepts to non-technical stakeholders

Experience leading technical teams or mentoring engineers.

Additional project-specific (visual similarity search) requirements:

Hands-on experience with production computer vision pipelines (not just research projects) - object detection, embedding models, vector/hybrid search

Familiarity with the ecosystem: DETR/YOLO-family models, Roboflow (or a comparable data management and training tool), vector databases (Pinecone or equivalent)

Ability to evaluate cost trade-offs between self-hosted GPU and managed/SaaS inference

Understanding of open-source model licensing constraints (AGPL, Apache, MIT) and their business implications

Experience designing systems with multiple parallel decision signals (multi-signal retrieval/ranking)

Working knowledge of AWS (ML/GPU area: SageMaker, EC2 GPU, Bedrock) - doesn't need to be a DevOps expert, but must be able to discuss architecture with the DevOps Engineer

Experience in making architectural decisions under time and budget pressure - the ability to say "this model is good enough, we don't need the latest SOTA" or "this is a licensing risk, we'll avoid it by choosing a different component," rather than defaulting to the technically most interesting but most expensive solution

Nice to have: prior experience in retail (product recognition, planograms, store display compliance).

What do we offer?

• •Contract of employment
• •Annual bonus
• •Private medical care
• •Cafeteria Platform/Multisport
• •English lessons subsidized by the company
• •Group insurance
• •Two additional days off (Good Friday, Friday after Corpus Christi) – with the possibility of exchanging for other holidays
• •Employee Referral Bonus Program
• •Attractive discounts for products and services at our stations
• •Employee stock purchase plan
• •Employee Assistance Program (Lyra)
• •Modern and convenient office that you can virtually visit here - https://goo.gl/maps/CLteHfYcdYMbdESq6
• •Trainings & possibility to develop skills in a wide international environment

Want to know even more about us? Take a look at our career page: https://workwithus.circlek.com/global/en/businesscentrewarsaw

Interested?

We encourage you to apply.

We know great companies are built from within, by great people like you. Come grow with us! We´re looking forward to your application.

When working with us you can depend upon it that you will not be judged on the grounds of race, national origin, gender, sexual orientation, disability, age, or other legally protected status. Oppositely – we believe that our diverse and inclusive culture helps us create an amazing atmosphere where everybody feels welcome.

Check who we are here: https://youtu.be/td-QGnNnvW0

We hereby inform that in the company Circle K Business Centre Poland sp. z o.o. with registered office in Warsaw an Internal Notification and F