Lead AI Product Engineer

The role

Cogentiq I2C is Fractal’s agentic AI Product for invoice-to-cash, covering Collections, Cash Application, Deductions, Invoice Management and Credit Risk. It runs on a Next.js front end, a FastAPI service layer, the Cogentiq agentic runtime, and a data tier on Azure Databricks with PostgreSQL.

You own the AI side: the agents, the models and everything that makes them accurate and dependable in front of a finance team. You design it, you lead the team that builds it, and you write code yourself.

What you will own

• Technical design of the AI side: agent decomposition, orchestration patterns, tool design, memory and context strategy.
• The agents across all five modules, from document extraction and matching through to the reasoning steps behind a recommendation.
• Accuracy and reliability. Evaluation sets, regression suites, and a defensible number for how well each agent performs before it ships.
• Guardrails and human oversight: confidence thresholds, escalation to a person, audit trails, and traceable reasoning for any decision touching cash.
• Predictive and machine learning models where they serve the product better than an agent does.
• Model strategy: choice of models, cost and latency per workflow, and the ability to switch providers without a rewrite.
• The interfaces between your side and the platform, agreed jointly with the engineering.

How you will work

• Lead the AI and agent engineering team. Set standards, coach engineers, and keep the team unblocked.
• Stay in the code. Take the hardest agents yourself and set the patterns others follow.
• Work as a pair with the Engineering Lead.
• Work with product management on what an agent can realistically be trusted to do, so external commitments are grounded.
• Be able to explain an agent’s behaviour to a finance stakeholder who will not accept "the model decided".

What you must have

• Ten to fourteen years in AI, machine learning or software, with at least three leading a team.
• Track record of owning technical design, not only implementing someone else’s design.
• Production experience with LLM and agentic systems: multi-step workflows, tool use, orchestration frameworks, and the failure modes that only appear at scale.
• Real evaluation discipline. Evidence that you have measured agent quality with something more rigorous than manual spot checks.
• Strong Python engineering. Your team ships production code, not notebooks.
• Document understanding and information extraction experience, ideally on messy real-world inputs such as emails, remittance advices and attachments.
• Classical machine learning depth alongside the generative work, and the judgement to know which problem needs which.
• Enough understanding of services, data and deployment to build agents that fit the platform and to hold your side of a design argument with the Engineering Lead.
• Product track record: AI built for many clients, not one-off delivery for a single engagement.

Good to have

• Order-to-cash or accounts receivable domain knowledge: collections, cash application, deductions, remittance handling.
• Experience with agent observability and tracing tooling.
• Experience getting AI through enterprise risk, compliance or model governance review.