• Lead engineer supporting large-scale post-training for frontier LLMs and emerging multimodal models.

• Partnered with research teams to train early-stage multimodal models combining vision, language, and action-based inputs, enabling experiments in grounded reasoning and embodied interaction.

• Developed training pipelines that integrate visual encoders, cross-attention fusion layers, and instruction-tuned language heads, optimizing for sample efficiency and multi-sensor alignment.

• Led RLHF workflow development for multimodal systems, including preference data generation for vision-language tasks (image reasoning, spatial queries, action prediction).

• Developed curriculum-based sampling and multimodal reward-model training strategies that improved robustness on tasks involving perception, visual QA, and trajectory evaluation.

• Designed distributed replay buffers and sampling policies for long-horizon action sequences, improving convergence in tasks involving temporal dependencies and sequential decision-making.

• Supported parallelism (TP/PP/DP) for large-scale multimodal and LLM training runs, focusing on reliability, reproducibility, and stable multi-view data loading.

• Collaborated on model merging, evaluation pipelines, and safety-alignment experiments for multimodal policy models.

• Collaborated with ML researchers to design, run, and analyze LLM training experiments, including supervised fine-tuning, instruction-tuning, and alignment evaluations.

• Built and maintained training pipelines for data preprocessing, tokenization, model configuration, hyperparameter search, validation checkpointing, and drift detection.

• Supported multi-round human-in-the-loop experiments: synthetic data generation, preference labeling workflows, reward-model retraining, and behavioral analysis of updated policies.

• Conducted rigorous analysis of model behavior in reasoning, multi-turn dialog, safety, and grounding tasks, producing reports that informed architecture and training-schedule changes.

• Contributed to early multimodal + embodied AI experiments, integrating LLMs with robotic simulators to study reasoning and control.

• Developed an internal web application to manage document requests, integrated APIs for request completion and database interactions, passing all internal security and quality assurance tests necessary for launch readiness.

• Implemented a growing portfolio of agentic robotic processing automations in Python

• Developed an internal Python package to simplify and accelerate robotic process automation development

• Train and support a growing group of developers

• Implemented multiple machine learning classifiers using scikit-learn and NLTK to replace a manual intent classification process

• Lead automation efforts as the business process owner surpassing an organizational goal of $2,000,000 in expense reductions

• Combined toolsets such as SQL Server, Oracle, and Tableau to deliver high-performance dashboards increasing the accessibility of data intelligence and analysis

• Analyzed prescribing habits leading to an increase in the utilization rate of a lifesaving drug by 19%, earning the organization an award for innovation

• Automated the updating of various tables in multiple SQL scripts using Python, contributing to an efficient and successful system upgrade

• Collaborated across technical and non-technical teams eliciting requirements and developing solutions for the implementation of an enterprise data warehouse

• Led ongoing fraud detection analysis that had an initial impact of $38,000 in savings

• Mentored, and trained colleagues on tools, data flows, and data warehousing