Alex ChandlerApplied AI & engineering
I’m Alex Chandler, an AI Scientist at BCG X. I work on AI evaluation, applied research, and software engineering. This site brings together some of the systems I’ve built, the research I’ve published, and projects I’ve pursued along the way.
Featured Work
Search & filter work ↗
Select a work type to explore projects.
Publications
& intellectual property.
Google Scholar ↗Patent application
Systems and methods for measuring performance of large language models
Explore ↗SemEval 2025Deloitte (Drocks) at SemEval-2025 Task 3: Fine-Grained Multilingual Hallucination Detection Using Internal LLM Weights
Explore ↗FinNLP 2025Deloitte (Drocks) at the Financial Misinformation Detection Challenge Task: Enhancing Misinformation Detection through Instruction-Tuned Models
Explore ↗EMNLP 2024Detecting Errors through Ensembling Prompts (DEEP): An End-to-End LLM Framework for Detecting Factual Errors
Explore ↗RL 2022Decision Transformer for Robot Imitation Learning
Explore ↗Education & experience.
Explore experience ↗Aug 2025–Present
BCG
Voice AI, evaluation, and enterprise systems.
Explore work ↗Jul 2024–Jul 2025Deloitte
Enterprise AI applications, backend engineering and DevOps.
Explore work ↗Jun–Aug 2023Fidelity
Financial question answering, document extraction and automated evaluation.
Explore work ↗Oct 2020–Mar 2022Keysight
Software engineering for network visibility products.
Explore work ↗2019–2024UT Austin
A foundation in computer science, artificial intelligence and machine learning.
Explore work ↗Building AI applications, evaluating their behavior,
and researching the methods behind them.
- Evaluation perspectivesFaithfulness check; factuality check; reasoning error check.
- AI inferenceEvaluate the same source and generated summary from each perspective.
- EnsembleCombine the prompt judgments.
- CalibrationEstimate a probability from the combined judgments.
- Inputs at each timestepState, previous action and return-to-go.
- Input representationConcatenate the inputs and add learned position embeddings.
- Causal transformerProcess the sequence of timesteps.
- Action distributionsPredict an action distribution for each timestep.
- Shared documentInclude the context once.
- Multiple questionsPackage the questions and instructions in one prompt.
- AI inferenceProduce the answers in one generation.
- Multiple answersEach indexed answer uses the shared context.