Work

Selected applied ML systems and research engineering work.

Applied ML

Industry systems

Production-oriented systems for foundation-model evaluation, multimodal data, and efficient inference.

Vector Institute · Foundation-model evaluation · 2026

Fine-Grained Benchmark Generation

An automated system for building grounded, technically demanding evaluation tasks from reference material.

Built OCR ingestion and a multi-agent designer/verifier workflow with repair, validation, and semantic deduplication.

Vector Institute · Multimodal data · 2025

Open-PMC-18M data pipeline

A large-scale extraction and preprocessing workflow for compound medical figures and their textual context.

Owned preprocessing and cut Qwen2.5-VL decoding from 4.0s to 1.2s and summary generation from 4.0s to 0.6s per sample.

Research engineering

Research systems and benchmarks

Post-training, interactive agents, robust evaluation, and visualization intelligence.

EACL 2026 Main · Post-training

RL-Text2Vis

A GRPO framework that evaluates text-to-visualization outputs after code execution.

Improves code executability and chart quality over strong prompting and supervised baselines.

EACL 2026 Findings · Multimodal agents

DashboardQA

A benchmark for agents answering questions through real interactive dashboards.

Co-created it to expose failures in visual grounding, planning, GUI interaction, and multi-view reasoning.

ACL 2025 Findings · Benchmarking

ChartQAPro

A challenging chart-question-answering benchmark designed around diverse real-world visualizations.

Co-created 1,341 charts and 1,948 questions used to evaluate 21 open- and closed-source VLMs.

ACL & EMNLP 2025 Industry Tracks · Evaluation

Large and tiny VLM judges

Reliable and efficient automated evaluation protocols for chart models.

A 2B judge matched a 7B baseline while reducing inference latency.

Foundation

Earlier research

EMNLP 2024 Main · Agentic generation

DataNarrative

An Actor–Critic-style workflow for planning, reflecting on, and revising data-driven stories with visualizations.

Introduced a benchmark of 1,449 real-world data stories for system and human evaluation.