Profile
I help teams turn uncertain technical questions into tested models, prototypes and software. My background spans graduate mathematics, undergraduate computer science, software engineering at NASA Ames and computational neuroscience. I developed a method for constructing shorter psychometric measures using machine learning and Shapley values, published in 2024. I move quickly between research and implementation, building the infrastructure needed to investigate a problem and test the result.
Consulting Focus
I take on feasibility and architecture investigations, applied AI prototypes, and fractional R&D engagements. The work can include retrieval and agent workflows, evaluation design, mathematical modelling and independent technical review. I work directly on implementation and can scope and coordinate specialist contributions where needed.
Selected Work
Nexus — AI agent operations and authorization
An operations platform where LLM agents research, plan and draft. The model cannot grant itself sending authority or directly change operational state.
- Built a typed operation registry and plan compiler, exposed through an MCP interface, to validate requests before execution. Human review binds approved content to immutable snapshots, content hashes and exact dispatch permits.
- Implemented scoped service keys, role checks and kill switches; append-only audit events and durable Neo4j queues with atomic claims, idempotency, retries and reconciliation.
- Built research and drafting workflows with identity checks, claim validation and human review in Python/FastAPI, with a Next.js operator console. Planning supports OpenAI and Anthropic adapters. Drafting failures produce templates or blocked generation; local drafting does not silently fall back to a paid API.
Correspondence Matrices — Boolean expression evaluation
Developed a computational architecture that represents Boolean functions as binary matrices indexed by truth assignments. Logical expressions can be combined through operations on aligned matrix representations.
- Built Python compilers around a shared symbolic intermediate representation, with eager and lazy construction and hybrid evaluation paths. Structural normalization, hashing and caching support reuse of compiled expressions.
- Built reproducible benchmarks with explicit truth-table checks outside timing windows, per-trial CSVs and consolidated reports. Documented compilation costs and evaluation timings separately so the implementation and measurements can be examined together.
Implementation and benchmarks · Working paper: Correspondence Matrices; Algorithms for Propositional Logic
Fractilate — mathematical art and encoded images
2019 – present · [fractilate.com](https://fractilate.com)
A platform in development for exploring equations, fractals, attractors and landscapes through visual work, intended to encourage interest in mathematics. Built primarily solo, with small components contracted out.
- Developed web, Python desktop and browser-studio implementations of a custom rendering core, with parity tests across implementations.
- Built composition scoring and preference learning, plus recoverable message encoding using a variant of the CM representation. A public decrypt tool and compatible Node/.NET encoding connect the mathematical work to a separate application.
Presidency or Prison — game simulation and development
2024 – present · [presidencyorprison.com](https://presidencyorprison.com)
A card game about truth, incentives and societal decisions, with software to design, simulate and test its rules.
- Built a Python simulation engine using NumPy, pandas and NetworkX, structured configuration and an operator CLI; developed a digital game in Dart.
- Built the supporting card-generation service, glyph geometry and validation, and video-production tools using FastAPI, Next.js, Playwright and FFmpeg.
Across Nexus, Fractilate and Presidency or Prison, research, media and workflow tools are infrastructure built to support the products.
B-Theory — mathematical research on self-awareness
A long-running independent research programme seeking a mathematical understanding of self-awareness: how a system can distinguish, represent and reflect on its own state. It draws on Matte-Blanco's distinction between symmetric unconscious relations and asymmetric conscious distinctions, exploring p-adic and ultrametric structures for the former.
- Algebraic Logic: develops Boolean modules and logical state-vectors over Boolean rings, providing the algebraic language for representing expressions and their measurement.
- Logical Topology: represents truth assignments as points and propositions as sets; explores inference through intersections and comparison through Hausdorff and truth-distance measures. A state unchanged by reflection—a fixed point—is a candidate model of self-recognition.
- Logical Homology: investigates chains and cycles of logical expressions and proposes gluing overlapping expressions into simplicial complexes, to study how logical relations could organize concepts and self-reference.
- Evolutionary Logic: explores logistic functions and Heaviside thresholding to connect continuous change with discrete truth values, including logical matrices evolving towards particular Boolean operators.
Correspondence Matrices originated here as a tool for formalizing conscious asymmetry, before becoming a separate computational project.
Experience
Independent R&D — product development and contractor direction
Built the products above and pursued independent mathematical research. Hired and technically directed dozens of freelance specialists across more than 100 engagements, including extensions. Scoped software, design, media and research work, with recurring part-time collaborations lasting multiple years.
NASA Ames Research Center — Software Engineering Lead
December 2022 – August 2025
Software engineering supporting Thermophysics Research.
- Developed AI-assisted document analysis, information extraction and knowledge-management workflows using local LLMs, NLP and retrieval.
- Built numerical software and integrated communications infrastructure, PLCs, instrumentation and high-speed data acquisition.
- Coordinated software and hardware integration, documentation and technical reviews supporting Critical Design Review and System Integration Review readiness.
Palo Alto University — Computational Neuroscientist
July 2018 – July 2022
- Designed and built neuroimaging processing and analysis pipelines in MATLAB and FSL. Co-authored the resting-state network study cited below.
- Developed a method using machine learning and Shapley values to select questionnaire items for shorter psychometric measures, evaluating them against full-scale score categories.
- Investigated how item selection changes with the intended measurement target. This method-development work informs my approach to AI evaluation: choose the target explicitly and check what the reported result actually measures.
Publications arising from this research
- Droncheff, B., Liu, K., & Warren, S. L. (2024). Item Retention as a Feature Selection Task: Developing Abbreviated Measures Using Shapley Values. Journal of Psychopathology and Behavioral Assessment, 46, 220–234. First author.
- Warren, S. L., Geller, W. N., Droncheff, B., & Liu, K. (2025). Individual differences in anxious apprehension and anxious arousal alter resting-state network structure and connectivity. Journal of Affective Disorders Reports, 20, 100895. Co-author.
Early Research
NASA Jet Propulsion Laboratory — 2003–2006; 2008–2009. Built LabVIEW drivers and I/O for physics instrumentation and laboratory automation; developed adaptive algorithms and scientific software in Java, C and MATLAB.
TU Berlin — Summer 2006. Developed C++ libraries supporting molecular beam epitaxy research.
California Institute of Technology — Summer 2004. Built computer-based experiments studying human rationality and decision-making.
Education
M.S. Mathematics — ALGANT Erasmus Mundus Program, University of Padova and Chennai Mathematical Institute, 2015. Dual-degree programme in algebra, geometry and number theory.
B.S. Computer Science, Minor in Mathematics — California State Polytechnic University, Pomona, 2011.
Selected Technical Skills
Languages: Python, TypeScript/JavaScript, C++, C#, Java, Dart, MATLAB, SQL, LabVIEW.
AI and systems: LLM integration, retrieval, agent authorization, evaluation and benchmarking; FastAPI, Next.js/React, PostgreSQL, Neo4j, Docker.
Scientific computing and verification: NumPy, SciPy, pandas, NetworkX, FSL; pytest, Vitest, Playwright.
Discuss a Project
For a scoped investigation, prototype or ongoing R&D engagement, email Brian with the problem, the constraints and the decision you need to make.