AI/Software
Achievements
Marcus’s presentation demoed how to leverage knowledge graphs to improve agentic reasoning and enhance retrieval-augmented generation (RAG) systems. His demo was featured in AI Tinkerer’s biweekly newsletter. It made the “top 5 demo” list!
AI Tinkerers is a global community of nearly 100,000 AI engineers and researchers across over 200 cities who build real systems, share unfinished work, and push the practical limits of modern models.
AI Agents
iconsult-mcp
AI Consultant for Multi-Agent Architecture
Python package (MIT License) 2026
A Model Context Protocol (MCP) server that serves as a technical architecture advisor for multi-agent systems. The server is built around a knowledge graph extracted from Agentic Architectural Patterns for Building Multi-Agent Systems, providing evidence-based recommendations grounded in published architectural patterns rather than generic advice.
Capabilities:
- Browse 141 concepts across 10 relationship types (462 total relationships)
- Explore pattern relationships, alternatives, and conflicts
- Search the source textbook with retrieval-augmented generation (includes page references)
- Get evidence-based architectural recommendations with citations
Installation:
pip install git+https://github.com/marcus-waldman/iconsult_mcp.gitlitrev-mcp
AI-Assisted Literature Review for Claude
Python package (MIT License) 2026
A Model Context Protocol (MCP) server that turns Claude into a research assistant for systematic literature reviews. The researcher stays in the loop throughout — directing searches, evaluating results, and making decisions — while Claude handles the retrieval, organization, and synthesis. It connects to Zotero, searches academic databases (PubMed, Semantic Scholar, ERIC), and supports building knowledge graphs to map the structure of a literature.
Capabilities:
- Search and retrieve papers across multiple academic databases
- Full Zotero integration for managing references and PDFs
- Citation snowballing (forward and backward)
- Semantic search over indexed papers using embeddings
- Argument mapping to build living knowledge graphs from literature
Installation:
pip install git+https://github.com/marcus-waldman/litrev-mcp.gitOptimization
arcopt
Adaptive Regularization using Cubics for R
R package (MIT License) 2026
An R package implementing Adaptive Regularization with Cubics (ARC), a robust nonlinear optimization method designed for ill-conditioned models where standard optimizers fail. The package automatically escapes saddle points and handles negative curvature—common challenges for maximum likelihood (or maximum-a-posteriori) estimation with mixture and nonlinear mixed-effect models.
Key features:
- Exact Hessian and quasi-Newton (gradient-only) optimization modes
- Cubic regularization that augments the standard quadratic approximation with a stabilization term
- Support for box constraints (lower and upper bounds on parameters)
- Built-in momentum acceleration for ill-conditioned problems
- 100% convergence on 80 benchmark test cases
Installation:
pak::pak("marcus-waldman/arcopt")