FDE Senior Engineer, AI & Capital Markets

KX•New York, NY

10d•$125,000 - $250,000•Hybrid

About The Position

This is a senior, hands-on delivery role with the Forward Deployed Engineering team at KX, functioning as part solution engineer and part AI engineer. You will work alongside quants, traders, e-trading desks, and research teams within their own environments, designing and building systems that integrate language models, agents, and vector search with the industry's deepest time-series data. Collaborating with innovative investment banks, hedge funds, market makers, and exchanges globally, you will own the delivery process from initial design through to production, with project timelines measured in weeks rather than quarters. The role requires sufficient domain knowledge to gain the trust of the desk and enough engineering depth to ensure systems operate at institutional volume. Forward Deployed Engineering at KX involves working with leading customers on their most challenging problems by embedding senior engineers within their environments. These engineers collaborate with customer teams (quants, traders, risk, platform engineers) to design, build, and prove solutions, remaining involved until the systems are running in production. The team is intentionally small, senior, and high-impact, with current efforts focused on establishing standards, architectural patterns, and operational methodologies, offering a unique opportunity to influence these foundational aspects.

Requirements

Working knowledge of embeddings and vector search: how to choose and tune an index, how to measure recall honestly, and why the obvious chunking strategy fails on financial documents.
Real data-modelling instinct: schema, partitioning and storage design for time-series at volume, and the judgement to know which decisions are expensive to reverse.
Comfort being the most technical person at the table and the most commercially aware person in engineering.
Python and SQL are essential.
The AI stack in practice: at least one major model API or an open-weights deployment, an orchestration or agent framework, and a vector store such as KDB.AI, FAISS, pgvector, Milvus or Qdrant. Considered opinions about what does and does not work matter more than brand familiarity.
Strong capital-markets experience building front-office systems (trading, pricing, market data, risk or research platforms) in production and under real market conditions.
Hands-on experience building and shipping LLM-backed systems: retrieval architecture, evaluation, tool use and the operational reality of running them once real users arrive.
Six or more years in or around capital markets technology: sell-side, buy-side, a fund, an exchange, or a vendor who served them well.
Genuine depth in at least two of: equities, FX or futures microstructure; derivatives and pricing; market or credit risk; trade lifecycle analytics; quantitative research workflows.
Real-time and time-series data engineering: tick stores, streaming, historical replay and intraday reconciliation at institutional volumes.
Consultative range: you can run a discovery workshop, handle a sceptical head of trading, and write a solution document that survives procurement.
Practical experience with the plumbing: FIX engines, market data handlers, exchange protocols, and the reference and corporate-actions data that decides whether anything reconciles.

Nice To Haves

Experience with q and kdb+ is highly desirable but not required. The appetite to become genuinely good at it matters more.
Equivalent evidence counts. Depth in another tick store or time-series engine (OneTick, ClickHouse, InfluxDB, Arctic or an in-house build), or in array and functional languages (APL, J, OCaml, Haskell, Scala) where the thinking transfers directly to q.

Responsibilities

Identify viable use cases: work with quants, traders and researchers to identify where AI genuinely changes the economics, and to rule out the cases where a well-written query would do. Feasibility, data readiness, cost to run, and an agreed definition of good enough before anything gets built.
Map and prepare the data estate: work alongside the customer's quants and data owners to understand the estate and its ontology, then join unstructured sources (research, filings, news, broker commentary, chat) to structured market and trade data. Point-in-time correctness, corporate actions, symbology and survivorship are what decide whether the answer is right.
Build the retrieval and inference layer: embedding pipelines, chunking suited to financial documents, index selection and tuning, and hybrid search across vector similarity plus time, symbol and entitlement filters. Then the serving path: latency budget, cost per query, caching, and behaviour under load.
Build the agent and tooling layer: give models controlled access to the customer's data and analytics through tool and function definitions, context assembly and multi-step orchestration, with guardrails including entitlements, so an agent can never see what the user cannot.
Define the evaluation framework: build the evaluation harness alongside the customer's own experts: golden sets, accuracy and recall measures, regression tests, and latency and cost benchmarks. A system that cannot be measured will not be approved for production.
Design the data platform: schema, partitioning, on-disk layout, attribute and index selection, compression and query paths, so the analytics the desk wants next year are straightforward rather than a rebuild.
Lead through to production: deploy into a secure, regulated environment, with monitoring, alerting, index and model refresh, a runbook and a documented handover. You remain engaged until the system is running in production.
Build reusable assets: reusable accelerators, reference implementations, and clear requirements back to KX Engineering, so the next engagement starts further forward than the last.