Software Engineer, Data Platform Lead (Horizon AI)

Horizon AI is building the operating system for how organizations run themselves. The first step is autonomous data centers. A fleet of AI agents senses the state of STN’s GPU infrastructure, reasons about what it finds, and increasingly acts on it. This is running in production against real fleet data, not a prototype. Agents already take action where the risk is bounded and the outcome is verifiable. Everything riskier arrives as a recommendation for a human, and the work is moving decisions from the second category into the first, one class at a time, on evidence rather than optimism.

Data centers are the proving ground, not the destination. We start here because infrastructure is measurable and unforgiving: an agent is either right, or the fleet tells you it was wrong. What we build here generalizes, and we know where it goes next.

Level: Principal (MTS 5) · Reports to: Chief AI Officer

The role. Build the data foundation the platform reasons over. Today our agents assemble a picture of the fleet by querying several separate systems, and we cannot prove wider autonomy because we do not systematically capture how humans decide. You fix both.

The problems you inherit

• Fleet state is scattered across a relational store, a DCIM system, and multiple time-series and monitoring systems. Each has its own query path. There is no unified model. - No decision-capture pipeline exists. We record what agents did, but not what operators saw, concluded, and chose. That record is the evidence base for granting agents wider autonomy.

• Our retrieval pipeline is document-oriented rather than fleet-state-oriented.

• We consume telemetry from systems other teams operate, where retention and granularity are tuned for dashboards rather than for learning from history.

You will

• Design and build the fleet knowledge model: assets, topology, configuration, and current state, queryable by an agent in one call rather than five

• Build the decision-capture pipeline, and instrument the operator workflow so we accumulate labelled human decisions from day one

• Own the data access layer that abstracts telemetry sources, so we can keep consuming today and build our own collection if that stops scaling

• Negotiate our requirements with the teams who own the telemetry: signals, granularity, retention, raw event access

• Own the scale path, including vector index behaviour and the analytics separation a growing schema will eventually need

You have

• Strong data modelling judgment, particularly modelling physical or infrastructure domains as something queryable

• Experience with time-series and event data at scale, including cardinality and retention trade-offs

• Depth in PostgreSQL, including vector search, and comfort with a large evolving schema - Enough infrastructure literacy to know what a fleet actually is: nodes, GPUs, fabric, schedulers, power. Familiarity with DCIM tooling is a strong plus

• A bias toward shipping a usable model early over designing a complete one

Nice to have: GPU fleet or HPC exposure, DCGM, Slurm, entity resolution across inconsistent sources