AI Engineer

ALTEN Romania is part of the ALTEN Group, Leader in IT and Engineering Consulting. We develop innovative and durable technical solutions that fulfill the needs of our local and international partners.

Role Summary:

We are looking for an AI Engineer to build the platform that reconstructs and validates software specifications from an existing system’s own assets — its source code, its existing documentation, and its existing test cases.

Working under the Technical Lead and within the architecture and quality standards they set, you will implement the analysis agents and pipeline stages that map components, draft faithful specifications, cross-check claims against their sources, and prepare concise, pre-validated review packages for subject-matter experts (SMEs). This is a hands-on engineering role focused on building reliable, evidence-based behaviour — not on setting direction.

The platform exists to accelerate SMEs, never to replace them. Your job is to make its outputs faithful, traceable, and trustworthy enough that an expert can adjudicate a short list of flagged items instead of reviewing everything from scratch.

Profile Summary:

This role is for a hands-on engineer who is energised by building trustworthy AI behaviour over real, existing systems. You take pride in outputs that are faithful, traceable, and honest about what they don’t know — and you understand that the goal is to make an expert’s review faster and sharper, never to replace their judgement.

Key Responsibilities:

1. Build the Analysis Pipeline

• Implement the agents and stages that map components, analyse them in depth, and draft specifications from source evidence.
• Decompose complex analysis into focused sub-tasks with clear inputs, outputs, and verification expectations, following the agreed architecture.
• Implement the verification pass that checks each claim back against its source.

2. Cross-Validation & Reconciliation

• Build the logic that reconstructs understanding independently from source code, existing documentation, and existing test cases.
• Implement comparison and reconciliation that flags inconsistencies, missing coverage, stale information, and likely defects.
• Apply the explicit rules that allow only low-risk, evidence-backed discrepancies to be resolved automatically, escalating the rest for human review.

3. Faithfulness, Traceability & Quality Outputs

• Implement and uphold the first-class outputs: claim-to-source traceability, confidence marking, gap register, discrepancy register, open decision list, and adjudication status.
• Ensure the system distinguishes implementation facts from inferred behaviour, and marks unsupported assumptions instead of promoting them to fact.
• Make sure gaps are recorded openly and never silently filled.

4. Evaluation, Testing & Reliability

• Write tests and contribute to evaluation harnesses that measure extraction accuracy, faithfulness, and run-to-run consistency.
• Debug pipeline failures and reconciliation edge cases, and improve the signal-to-noise ratio of what reaches the SME.
• Build for graceful failure and clear, traceable outputs rather than confident error.

5. SME-Ready Output & Collaboration

• Implement the concise, high-value review packages SMEs adjudicate, keeping the human gate mandatory before completion.
• Work closely with the Technical Lead and, where relevant, with SMEs to validate that outputs are faithful and reviewable.
• Raise design concerns and propose improvements within the direction set by the Technical Lead.

Required Qualifications:

• Solid software engineering fundamentals: clean, testable code and disciplined debugging.
• Practical experience building AI- or NLP-powered features — information extraction, document analysis, agent behaviour, or similar — beyond simple prompt-and-response usage.
• Ability to read and understand existing codebases and reason about components, dependencies, interfaces, and behaviour.
• Comfort with prompt and agent design, and with evaluating and improving the quality of generated outputs.
• Attention to evidence and provenance: a habit of tracing claims to sources and marking uncertainty rather than guessing.
• Ability to work effectively under technical direction and within defined quality gates.
• Clear written communication for technical notes, test results, and decisions.
• A reliable, detail-oriented ownership mindset over the components you build.

🚀Only eligible candidates will be contacted in a maximum of 2 weeks. Any information you provide is confidential.