QuantAI - Agentic Quant Manager

Job description

QuantAI Manager, Agentic Quant Practice: QuantAI, Accenture Commodity Markets (ACM) Level: Manager (L7) Locations: London, UK The roleQuantAI is a global team within Accenture's global Commodity Markets practice. We build decision systems around artificial intelligence (AI) for commodity and energy trading: power, gas, liquefied natural gas (LNG), oil products, and agriculture, plus adjacent financial markets. We are hiring a quant who sets the modeling bar: someone who has built models that traders, hedgers, or risk managers acted on, and who already does that work through coding agents every day. We are looking for a quant who is all three of these at once: •       Thinker: frames ambiguous market problems, chooses the right method, and knows when a result is too good to be true. •       Doer: builds the work hands-on with coding agents and ships evaluated models and tools, not just slides or notebooks. •       Communicator: explains models, risks, and trade-offs clearly and persuasively, whether to a trading desk, a client executive team, or a conference audience. This is a hands-on player-coach seat. You will design, build, and evaluate forecasting, pricing, optimization, and decision models, write the quant tools that agents call, and decide with evidence whether a new model actually beats a strong classical baseline. You will also lead client project teams and go-to-market initiatives, run hybrid agentic and human reviews of our codebases, coach others, and help decide which prototypes become repeatable client offerings. You will write code, direct coding agents, review their output, and take responsibility for the model logic and evaluation evidence. What you'd work on•       Choose methods across statistics, econometrics, mathematical optimization, scenario modeling, geospatial science, weather and climate models, machine learning (for example, boosting), deep learning, foundation model fine-tuning and training, and decision models. Connect the quantitative work to agent harnesses, reusable agent skills, and Model Context Protocol (MCP) servers. Choose the simplest method that meets the evaluation requirements. •       Build and extend our forecasting stack, which combines statistical baselines, fine-tuned time-series foundation models, and multi-model pooling behind MCP tools, with leakage-safe walk-forward backtests and probabilistic metrics. Design recursive self-improvement (RSI) loops in which agents propose, test, and keep or revert changes to foundation model tuning using validation data, with untouched test data reserved for final evaluation. •       Turn forecasts into decisions: formulate and solve optimization problems (for example, a gas-purchase mixed-integer program, and candidate demonstrators for ancillary-service offers, congestion-rights auctions, and margin and collateral planning) and wire them into agent workflows that traders can use. •       Build and evaluate decision models for text-heavy problems such as counterparty screening and news signals. Compare approaches using encoder classifiers, natural language inference classifiers, small language models, and large language models (LLMs), with calibration, deferral to human review, and held-out evaluation. •       Write quant code designed to be called by agents: typed inputs, explicit units and trading calendars, reproducible calculations, controlled randomness where needed, and tests that expose errors. Package it as MCP tools and reusable agent skills. •       Own the quant evaluation gate inside our agent harnesses, so that agent-written research passes checks for leakage, baseline parity, units, calendars, and out-of-sample performance before anyone trusts it. •       Run hybrid agentic and human codebase reviews: use review agents to sweep code for defects, leakage, and test gaps, then apply human judgment to model logic, market conventions, and what is ready to ship. •       Lead client project teams of domain experts, quants, full-stack engineers, and analysts from problem framing through delivery: sit with client trading, origination, and risk teams, turn loosely framed problems into working, evaluated assets quickly, and explain to senior stakeholders what a model can and cannot be trusted to do. •       Lead go-to-market initiatives with practice leadership: shape offerings, build demonstrators, contribute to proposals, and present to client trading and risk leaders. •       Coach teammates and set review standards for quantitative work across the team. Early priorities could include taking a trading demonstrator from prototype to a backtested result that meets agreed evaluation requirements; moving a decision model prototype (for example, counterparty screening) from synthetic benchmarks onto a realistic held-out evaluation; and adding commodity-specific checks to the shared research and verification workflows that our quants, full-stack engineers, and analysts build on. What good looks likeTreat the Must-have list below as the minimum screening bar for the role. The Nice-to-have list is additive, not required. Must-have•       Bachelor's degree in mathematics, statistics, physics, engineering, computer science, economics, finance, operations research, or a related quantitative field. An associate degree is acceptable with at least 2 additional years of directly relevant experience and a strong record of applied quantitative work. •       Minimum 5 years of experience in a quantitative role where your models fed trading, hedging, pricing, or risk decisions: for example on a commodity, energy, rates, foreign exchange (FX), credit, equity, options, or crypto desk, or in a systematic trading, structuring, or front-office strategy team. •       Hands-on experience in at least one of: derivatives pricing and hedging, forward-curve or volatility modeling, price or demand forecasting, portfolio or asset optimization, or systematic strategy research. •       Rigorous research practice you can show: walk-forward or out-of-sample backtests against explicit baselines, leakage control, and honest reporting of results that did not hold up. •       Hands-on experience formulating and solving mathematical optimization problems (for example, linear, mixed-integer, stochastic, or robust optimization) tied to commercial or operational decisions. •       Expert Python for quantitative work, including tabular data processing, high-performance central processing unit (CPU) and graphics processing unit (GPU) compute, optimization solvers, and statistical, machine learning, deep learning, foundation model, generative modeling, and geospatial packages, with Git-based collaboration, tests, and code review. •       Current, daily practice building with coding agents such as Claude Code or Codex: you direct agents to write, test, and refactor research code, you catch and fix what they get wrong, and you have built or extended at least one agent workflow, tool, or skill that other people or agents use. •       Hands-on experience applying modern model classes to quantitative problems, including machine learning, deep learning, foundation models, LLMs, small language models, encoder classifiers, and decision models. You have fine-tuned or adapted at least one of them and measured whether the gain held up out of sample against a classical baseline. •       Evidence that you can lead technically: set direction on a problem, lead a team or workstream, and review others' work rigorously (including agent-written code). •       Evidence of strong communication: you have presented quantitative work to traders, clients, or senior stakeholders and can explain model logic and its limits clearly and persuasively, in writing and in person. Nice-to-have•       Commodity or energy trading experience, especially power, gas, LNG, oil products, freight, or agriculture, including physical constraints such as storage, transport, delivery calendars, and settlement conventions. •       Master's or P