Company: Nyxium
Location: London
Posted: May 9th, 2026
Title: Power Systems Consultant (Part-time, 10 hrs/week)
Compensation: $150 per hour
Objective: Encode power feasibility, timeline, and risk into Nyxium’s decision engine
Nyxium is a London-based deep tech company backed by top-tier investors, building an infrastructure intelligence platform for siting and deployability decisions across data centers and energy infrastructure.
We are seeking a Senior Power Systems Consultant to strengthen our grid access predictive capability, perform technical modeling, and elevate our grid intelligence beyond basic GIS mapping. The work will be conducted over six weeks, with clear deadlines and deliverables to ensure focus and impact within Nyxium’s ecosystem.
Define a rule-based system that enables Nyxium to reliably determine whether a site can obtain power, on what timeline, and what constraints may prevent it. The consultant will define the rules once, and Nyxium will apply them automatically across thousands of sites using geospatial data and scoring logic.
You will define indicators and heuristics for grid interconnection that Nyxium can compute as proxies for interconnection feasibility. The work will begin with data centers, and the methodology will be expanded in week 4 to additional markets, including battery energy storage and renewables. These proxies will be used to evaluate whether a site can be connected to the grid. They must include indicators of what makes a site power-viable versus non-viable, integrating key constraints such as:
These indicators and heuristic rules will be used to construct a power readiness score, with components that reflect structurally favorable or unfavorable conditions for power supply or grid connection at specific sites, such as indicative spare capacity. They should also propose hard no-go rules that depend on the asset class, for example, “> X km from ≥132 kV → NO for ≥50 MW data centers,” as well as upgrade classes (minor works versus transmission upgrades required and/or new substation build-out). In addition, they should specify, where appropriate, the required inputs Nyxium must collect.
On week 2, you will expand upon the first week’s methodology to translate real-world grid behavior into usable timelines. You will define:
With this work, we should be able to produce a structured mapping from site characteristics to timeline estimates. The mapping rules (if/then) may be based on factors such as voltage level, region (US/EU), upgrade class, load density, renewable and data center density, and congestion proxies. An estimate of the confidence associated with each proxy and approach must also be provided (e.g., high / medium / low), depending on data caveats and other well-justified dimensions.
By week 3, the methodologies should be safeguarded against hidden blockers. Based on the proxy models developed, critical risk patterns and common failure cases must be well documented. You will define:
In week 4, the models will be reused and expanded to cover battery energy storage systems and renewable energy assets. Indicators and heuristics for the co-location of data centers with these assets will also be developed. The outputs for this week will be:
Novel adjusted weights for:
You will review real site outputs from Nyxium and answer:
You will receive feedback that directly improves the model, validate 15–20 real sites, and, for each, provide a score, upgrade and timeline classifications, an explanation of the decision (e.g., would you proceed? — Yes/No/Conditional, with justification), and a discrepancy report (for sites deemed infeasible, explain what could make them viable, where appropriate).
Consultant must provide:
By the end of week 6, the Power Readiness Score and timeline classes must be live in the product. Greater than 80% agreement between consultant judgment and the model across 10 validation sites must be demonstrated, and the decision output must be enabled as: Deployable / Conditional / Not Deployable (power-driven).
The methodology and code must be clearly documented, transparent, and reproducible, with a Python-based approach preferred and aligned with industry best practices.
#J-18808-Ljbffr