EPRI Technology Radar: Methodology & Platform
My role
I functioned as the data architect for EPRI’s Technology Radar, designing the statistical methodology, building the Python aggregation tooling, and authoring the backend, front-end, and methods documentation behind the institute’s flagship technology-foresight product. It’s the work I’m proudest of from my time at EPRI: a system that had to be both statistically rigorous and trusted by the experts whose judgment it encoded.
The hard problem
Technology foresight runs on expert opinion, and expert opinion is uneven. Some assessments come from deep specialists with high confidence; others are educated guesses. Treat them all equally and your scores are noise dressed up as signal. The central challenge was turning subjective, variably-confident input from roughly 80 subject-matter experts into scores that decision-makers could defend when allocating R&D budgets.
What I built
Confidence-weighted methodology. I evaluated competing weighting schemes — linear, exponential, Bayesian model averaging, and a Delphi variant — and selected a confidence-based inverse-weighting approach, so that higher-confidence expert input drives the aggregate score and low-confidence input is damped rather than discarded. The result lets genuine expertise carry more signal without silencing anyone.
Uncertainty quantification. Every technology is scored across 7 assessment parameters — Technology Readiness Level, Market Disruption Potential, Economic Impact, Sociopolitical Obstacles, Time to Commercialization, Environmental Impact, and Ease of Implementation — with variance, standard error, and confidence intervals computed per technology per parameter, then normalized to a 0–1 scale for cross-parameter comparison.
The Python tooling. I wrote the aggregation and analysis compendium: a converter that parses the proprietary expert-scorecard export and computes per-technology statistics, and a confidence-weighted sensitivity tool that compares weighting schemes across all parameters.
Driving the methodological upgrade. The Q4 2024 Pulse Report scored 48 technologies using simple crowd-sourced averaging. For the Q1 2025 report, I introduced the confidence-weighting methodology and helped expand coverage to 69 technologies — the upgrade that took the Radar from a poll to a defensible analytical instrument. I co-authored the resulting Pulse Reports, leading the statistics, methodology, and figures.
Why it matters
Energy R&D portfolios involve enormous bets under deep uncertainty, and the cost of chasing hype instead of substance is measured in years and millions. A rigorous, confidence-aware foresight system helps decision-makers separate genuine signal from enthusiasm — and put resources where they’ll actually move the energy transition forward.