We are seeking an experienced and visionary Software Architect with proven, hands-on experience in Generative AI (GenAI) to join our team.
In this key role, you will be responsible for designing high-level software structures, integrating AI-driven capabilities, making critical technical decisions, and ensuring the architectural integrity and innovation of our software solutions.
What you will actually do:
Own the technical arc of an engagement
Run discovery with business and technical stakeholders and convert vague ambition ("we want AI in our support flow") into a scoped, measurable, buildable system — or into a defensible recommendation not to build it.
Say no to use cases that AI will not serve well. Deciding what not to build is a first-class deliverable here, and you will occasionally have to say it to the person signing the contract.
Partner with sales on technical pre-sales: solution shaping, effort and cost estimation, architecture defense in front of a customer's CTO or security board.
Design AI systems that survive contact with production
Architect agentic systems end to end: task decomposition, tool and function calling, orchestration and hand-off patterns, state and memory, human-in-the-loop checkpoints, failure containment and recovery.
Engineer context, not just prompts — retrieval strategy, chunking and indexing decisions, hybrid and graph retrieval, re-ranking, caching, context window budgeting, and the data pipelines that keep all of it fresh.
Make and defend the model-portfolio call: frontier vs. small vs. open-weight, hosted vs. self-hosted, when fine-tuning or distillation actually beats better retrieval, and how to stay swappable as the frontier moves every quarter.
Design for interoperability using emerging agent standards (MCP, A2A and successors) rather than locking a customer into one vendor's orchestration layer.
Make quality measurable
Stand up evaluation harnesses as a condition of delivery: golden datasets, LLM-as-judge with human calibration, regression suites in CI, and online evaluation against real traffic.
Define the acceptance criteria that let a customer's team keep shipping safely after handover. If they cannot tell whether a change made the system better or worse, you have not finished.
Own the non-functional reality
Unit economics: cost per task, latency envelopes, throughput, degradation strategy. Systems that are correct and unaffordable are failures.
AI security: prompt injection and indirect injection, tool-use blast radius, data exfiltration paths, tenancy isolation, secrets handling, supply chain of models and MCP servers.
Governance and compliance: EU AI Act obligations (Article 50 transparency duties began applying 2 August 2026, with the high-risk regime following), data residency, PII handling, auditability, and the documentation a regulator or enterprise risk function will actually ask for.
Raise the bar around you
Mentor engineers and architects across delivery teams; run architecture and code reviews.
Codify what works into CodeValue reference architectures, accelerators, and internal IP so the next engagement starts further along than the last one.
Represent CodeValue externally — meetups, conferences, technical writing.
Stated plainly, to save everyone's time:
Architects who have integrated an LLM API but never owned the system's quality over time.
People whose GenAI experience is entirely prompt-level, with no view into retrieval, evaluation, cost, or security.
Anyone who believes the answer to every problem is a larger model or another agent in the graph.
Architects who produce diagrams and leave before the system meets production traffic.
