Team
MoA+
Project Concept
The Mixture of Agents (MoA) was proposed by Together AI (https://arxiv.org/pdf/2406.04692v1). This approach has been used for synthetic data generation, LLM-as-Judges, and, etc. A follow up of this work, https://arxiv.org/pdf/2502.00674v1, has somewhat simplified the approach. I believe we can potentially improve this approach (and maybe call this MoA+ or something else).
Entry
Status: Submitted
Last saved: May 18 at 1:47 PM PDT
Team Roster
Message board not available for this team yet.
Arindam Khaled Team Lead RSVP Approved
Applied scientist at Seekr
Had the original idea for an innovation on MoA and was the team lead for the project. Co-created the MoCA architecture, co-created the benchmark validation strategy, co-created the codebase. We had a team of only two, and we each participated in every aspect of the project.
Professional with a Ph.D. in Computer Science and extensive experience in machine learning, natural language processing (NLP), and optimal decision-making under uncertainty. Proven track record in developing innovative algorithms, optimizing (classical and deep learning) models, and enhancing user engagement.
Machine learning, natural language processing (NLP), optimal decision-making under uncertainty
Open source GraphRAG using BAML, Synthetic data generation, Confidence learning
Jordan Woltjer RSVP Approved
Economist at US Army Corps
Co-created the MoCA architecture, co-created the benchmark validation strategy, co-created the codebase. We had a team of only two, and we each participated in every aspect of the project.
Jordan Woltjer is an economist at the U.S. Army Corps of Engineers’ Institute for Water Resources, where he works on flood and navigation business lines. Holding an M.S. in Economics and Computer Science from Duke University, he leads projects on group-dependent confident learning, multi-hazard flood-damage modeling, and open-source data cleaning. Jordan thrives on turning messy public datasets into actionable insights and is passionate about applying AI for real-world resilience.
Areas of interest include label-noise theory; flood-risk modeling; econometrics and causal inference; and AI for climate adaptation and resilience.
Group-Dependent Confident Learning: recalibrates model training to accommodate mislabeled data with group- and true-label–dependent mislabeling probabilities.
Multi-Hazard Flood-Damage Modeling: builds models that predict structure and contents damage during flood events.
Flood-Data Enhancement Toolkit: standardizes disparate flood-loss and hydrologic datasets into a clean, rich resource for flood researchers and planners.