Principal Architect, AI & High Performance Systems

KX•New York, NY

7h•$150,000 - $350,000•Hybrid

About The Position

This is a principal-level, hands-on architect role with the Forward Deployed Engineering Team at KX. You assess customer environments and design AI-enabled, high-performance data and analytics systems, deciding where KX can create material value across AI, real-time data processing, accelerated computing and infrastructure modernisation. Working with the most innovative Investment Banks, Hedge Funds Market Makers and exchanges globally, you define how KX and their wider technology estate should work together in this new era of technology disruption. That means understanding their estate properly before proposing anything, being specific about what will get faster or cheaper and by how much, and then staying with it through to the first production outcome. You are the technical authority in the room, and the credibility of the architecture is yours to defend. Forward Deployed Engineering is how KX works with its leading customers on their hardest problems. Rather than handing a customer software, we put senior engineers inside their environment, alongside their quants, traders, risk teams and platform engineers, to design, build and prove the thing that solves the problem, and we stay until it runs in production. The team is deliberately small, senior, high impact and it is being built now, which means the standards, the architecture patterns and the way we work are still open questions. If you would rather set them than inherit them, this is the moment to join.

Requirements

Deep understanding of distributed systems, real-time processing, streaming architectures and system performance: you know why something is slow, not just that it is.
Strong cloud architecture skills and strong Python.
A working understanding of AI and agent-system architecture: model integration, tools, context, and how enterprise data actually gets to a model safely.
Credibility with CTOs, Chief Data and AI Officers, platform leaders and senior engineers. You can hold all four conversations in the same week.
Strong capital-markets experience building front-office systems (trading, pricing, market data, risk or research platforms) in production and under real market conditions.
A track record designing and delivering production AI, data-platform or high-performance computing systems at scale.
Practical experience with GPU or other accelerated-compute technologies.
Experience designing evaluation frameworks for AI systems: agreeing what correct means with domain experts, building golden sets, and measuring accuracy, recall, latency and cost honestly enough to support a deployment decision.
Experience bench-marking workloads and diagnosing performance, scalability and cost.
Experience operating in complex, secure or regulated enterprise environments, and the patience that requires.
Equivalent evidence counts. Low-level systems programming in C++, Rust or C, and the performance instincts that come with it: memory layout, cache behaviour, lock-free concurrency, kernel-bypass or FPGA work.
Depth in another time-series or tick store (OneTick, ClickHouse, InfluxDB, Arctic or an in-house engine), or in array and functional languages (APL, J, OCaml, Haskell, Scala) where the thinking transfers directly to q.

Nice To Haves

Experience with q, kdb+ is highly desirable but not essential. Welcome if you have it; the appetite to become genuinely good at it matters more.
Cloud Certifications desirable but not essential.

Responsibilities

Assess the environment: run architecture, AI-readiness and workload-performance diagnostics inside the customer's environment, and turn the findings into a defensible design.
Identify accelerated-compute opportunities: assess workloads for GPU and accelerated-compute opportunities, and prove the case with numbers.
Evaluate AI and agent use cases: feasibility, architecture, cost and what it takes to operate them. Saying “not yet, and here's why” is part of the job.
Design across the estate: one coherent system spanning KX technology, AI platforms and the customer's own data, not four loosely-joined ones.
Map the data estate: work alongside the customer's quants, data owners and domain specialists to understand the estate and its ontology, and the opportunities to leverage that data.
Quantify the gains: size the gains in latency, throughput, scalability and cost credibly enough to survive the customer's own engineers.
Define the engagement hypothesis: define the architecture and what each engagement is trying to prove, and how we will know.
Benchmark and prototype: benchmarks and rapid prototypes on real customer data, with results you would publish.
Define the evaluation framework: agree how each AI or agent system will be measured before it is built: golden sets, accuracy and recall, regression tests, and latency and cost benchmarks. Architecture that cannot be evaluated cannot be defended.
Lead through to production: senior technical leadership through the first production outcome, not a slide deck and a handover.
Build reusable assets: reusable diagnostics, reference architectures and accelerators so each engagement starts further forward than the last.
Inform the product roadmap: turn repeated customer requirements into clear inputs for Product and Engineering.