Explore the work.

Voice AI in production

Engineering a healthcare voice agent with real-time scheduling integrations.

Voice-agent simulation and evaluation

A simulation harness for realistic caller personas, automated scoring and behavioral regression tests.

Measuring the performance of large language models

Systems and methods for measuring performance of large language models.

AI Evaluation Tool

Assessment, verification and correction workflows for generated reports.

Agentic AI for government

Backend and DevOps engineering for an agentic AI application deployed for government use, serving 40K+ users.

Fine-tuning LLMs for batch prompting

LLM Efficiency Research

Fine-tuned language models that answer batches of questions in one inference pass.

Financial misinformation detection

AI Evaluation Research

Instruction-tuning for financial claim classification and reasoned explanations.

Finding hallucinations across languages

AI Evaluation Research

Token-level detection of hallucinated spans using features from internal LLM layers.

Mastering Leduc Hold’em

Reinforcement Learning Research

Comparing reinforcement learning and counterfactual regret minimization in imperfect-information poker.

Detecting factual errors with DEEP

AI Evaluation Research

Ensembling prompt judgments to detect factual inconsistencies in generated summaries.

Factuality detection with AlignScore and LLMs

AI Evaluation Research

Context decomposition and prompt-based evaluation of factual consistency.

Raft with RAPID

Distributed Computing Research

Integrating stable membership views into a consensus system.

Shared context. Multiple answers.

LLM Efficiency Research

Give a language model one document and several questions or requested outputs. Return the answers together in one inference pass.

Financial question answering system

Engineering a financial chatbot with named-entity recognition, table extraction and automated answer evaluation.

Temporal grounding in egocentric video

Computer Vision Research

Contrastive learning, hard negatives and temporal reordering for video-language representations.

Learning from mixed-quality demonstrations

Reinforcement Learning Research

Reformulating robot imitation learning as a sequence of returns, states, and actions.

Zero-shot controllable text generation

Language Model Research

Topical control through contrasting expert language models and self-organizing structures.

Network visibility engineering

APIs, test automation and AI threat detection for network visibility products.

AI red teaming for question answering

AI Safety Evaluation

Testing question-answering robustness by using modified beam search to find adversarial word sequences that reduce ELECTRA's accuracy.

Applied computer vision for musical instruments

Applied Computer Vision

Computer vision models for instrument identification and valuation, supported by a database of 10,000+ instruments.

Deep reinforcement learning for wells

Reinforcement Learning Research

Planning well paths through 3D subsurface data with actor-critic methods.

Geospatial analytics and backend engineering

Geospatial Analytics & Software Engineering

Analyzing political districts with graph theory and computational geometry; leading backend, API development and data collection for the GerryMap web application.