AI engineering built for the trading desk.

Blackdown designs and delivers governed AI systems for power trading, renewables origination and route-to-market operations — with deterministic controls and named human approval built in.

See the work

Why Blackdown

Front-office context, carried through to production.

Blackdown was founded on trading and risk experience at JPMorgan, followed by hands-on engineering inside a major European utility. Delivery experience includes reserve-service integration, GoO trade PnL attribution and BESS dispatch optimisation — systems that have to survive contact with registries, settlement and the Balancing Mechanism.

Production delivery across European power trading and renewable energy operations

PPAGuarantees of originRoute-to-marketBESS dispatchBalancing MechanismSettlement

What we build

Origination and pricing

AI-assisted offer drafting with economics reconciled against desk-approved pricing logic. Curves, volume profiles, indexation and certificate assumptions are surfaced for review.

Certificates and portfolio operations

GoO and REGO inventory matching across registries, vintages and delivery periods, with the resulting nominations and settlement evidence prepared for review.

Governed trading workflows

Multi-step trading and operations workflows decomposed into auditable decisions, deterministic services and explicit approval gates.

Featured work

Five process lines, each irregular until it reaches a single amber approval point, then straight

A German power trading firm

Governed AI workflows for PPA, GoO and route-to-market operations.

Request triage reduced from ~4 hours to ~20 minutes in production

Requests were decomposed into traceable steps, with deterministic processing and explicit approval before any binding action.

A draft pricing line weaving around a straight reference line inside a narrow tolerance band, with the two out-of-band points marked in amber

A European energy firm

AI-drafted offers an originator can verify and stand behind.

Against 24 historical deals, 78% reconciled to analyst pricing within ±3%

The control layer reconciled draft economics against desk reference pricing and surfaced exceptions for analyst review.

A price line spikes inside a highlighted window while a heavier state-of-charge line holds level through it; an amber segment beneath marks the reserved headroom explaining the hold

A major European utility

Governed AI decision support for BESS dispatch and revenue stacking.

Measured increase in BESS revenues in client operations

Deterministic BESS optimisation and reserve-service integration, with governed AI explanation that makes dispatch decisions commercially legible.

View all work

How it works

Defined scope. Clear milestones. Working software at every gate.

  1. 01 — Scope

    A short paid discovery defines the workflow, evidence base, principal risks and measurable outcome.

  2. 02 — Design

    Architecture, tool boundaries, control model and evaluation plan are agreed before build.

  3. 03 — Build

    Working software is delivered against milestones and reviewed with the people who will operate it.

  4. 04 — Evaluate

    Performance is assessed against representative historical data and the desk’s existing pricing or control logic.

  5. 05 — Hand over

    Production deployment, documentation, operating controls and knowledge transfer.

Explore how we deliver

Control comes before autonomy.

Models are useful for interpreting contracts, matching certificates and preparing decisions. They should not receive unrestricted authority to submit bids, move funds or change positions. Blackdown separates model reasoning, deterministic calculation and controlled execution, with named approval before any binding action.

FAQ

Frequently asked questions

Not without an explicit control design. Binding actions are routed through named approval gates before execution; the model itself is not given unrestricted authority.

Yes. We design within the client’s security, data-governance and model-risk requirements, and agree the approved tooling, data boundaries and deployment model before build.

Engagements are usually fixed-price and milestone-led once the scope is defined. Discovery is short and paid; the resulting statement of work sets out the outcomes, assumptions and acceptance criteria.

Not publicly. Our client work is anonymised by agreement. We can discuss relevant delivery detail and references under NDA once there is a serious opportunity.

Start with a working session on one specific workflow. If there is a credible fit, we propose a short paid discovery with a defined output and decision point.

Bring us the workflow that still needs manual checking.

We will show you where a model helps, where deterministic controls belong and where a named person must remain accountable.