Edinburgh·London·Cape Town

Electrical Engineering & Computer Science — University of Edinburgh

David Levy

Featured research

Bayesian Teaching for Personality Inference in LLMs

Six open-weight LLMs were taught to perform Bayesian belief updating over Big Five personality traits — extending the Qiu et al. (2026) framework from task-level preferences to stable trait inference.

Fine-tuned models matched the Bayesian teacher within 0.5pp on a 50-item questionnaire, and the inferred personality model transferred to an unseen flight-recommendation task with no task-specific training data.

6
open-weight LLMs fine-tuned
91.6–96.3%
teacher agreement
Llama / Qwen / Gemma
3B – 14B parameters

Try it · the Bayesian update step

1
2
3
4
5
Belief over Openness · 0/4 answereduniform prior

Q1.I have a vivid imagination.

Q2.I'm full of ideas.

Q3.I enjoy thinking about abstract concepts.

Q4.I'm interested in many different things.

1 = strongly disagree · 5 = strongly agree

Experience

Bloomberg

Data Engineering Intern

2025

Built automated Python pipelines integrating multiple internal sources, reducing processing time by 98%. Anomaly-detection logic surfaced data-quality issues feeding London-wide trading and risk reporting.

Full Spectrum

Operations Manager

2024–25

HealthTech startup supporting parents of autistic children. Drove product development and user research, ran user interviews, iterated on prototypes, and built investor pitch materials.

Solenya

AI Researcher

2024

Investigated zero-shot recommendation systems, boosting new-user recommendation accuracy by 15%. Synthesised findings into actionable technical reports.

Addionics

Data Analyst

2023

Automated data collection, cleaning, and visualisation of battery test results in Python, streamlining reporting for R&D decision-making.

Edinburgh AI Society

Funding Lead

2023 – present

Scaled membership from 30 to 400+. Secured Anthropic as headline sponsor, organised the Edinburgh AI Expo and recurring technical talks (100+ attendees per session).

See what I'm building
Platform

QuantBase

An open platform making finance intuitive for technical people.

Five interactive modules. Built from first principles.

Built with Next.js, React, TypeScript, Three.js, and Python

01

Glossary

Learn the language of finance.

02

Market Data

Real-time charts and volume.

03

Technical Indicators

RSI, MACD, Bollinger Bands, moving averages.

04

Options Pricing

Black-Scholes visualiser.

05

Backtester

Test strategies on historical data.