Foreign policy · Multilateral affairs · Evidence-grounded AI for policy work
Former United Nations political affairs officer, based in New York. Columbia SIPA and Sciences Po alumnus. My work connects political analysis, institutional knowledge, and software engineering.
I am building TACITUS and PRAXIS to help analysts move from source material to defensible judgment: preserving attribution, chronology, uncertainty, review decisions, and the evidence behind an output.
Architecture illustration · fictional policy scenario · Explore the layers and current capabilities
| Project | Purpose | Current status |
|---|---|---|
| PRAXIS | Analyst-facing workbench for documents, evidence, context capsules, agentic workflows, and inspectable outputs. | Active product development |
| DIALECTICA | Rust-first engine for building, validating, reviewing, and exporting portable, source-grounded PRAXIS Capsules. | Active public engine |
| Tacitus Policy Workbench | EdgeQuake-based, local-first integration lab for document ingestion, knowledge graphs, policy context, review, and capsule portability. | Prototype; end-to-end runtime verification pending |
| TCGC | Benchmark and corpus for evaluating whether AI systems preserve actors, evidence, commitments, contradictions, time, and provenance in policy and conflict texts. | Research build; measured leaderboard pending |
| TACITUS Knowledge Pipeline | Reference notebooks for turning source documents into evidence-linked knowledge graphs for policy and conflict analysis. | Public reference implementation |
See the public project status map for the distinction between active product work, research modules, reference implementations, and completed experiments.
AGON explores source-grounded extraction of actors, claims, commitments, contradictions, and friction patterns. KAIROS explores temporal knowledge graphs for events, commitments, episodes, and institutional change. Both are experimental research systems rather than production decision tools.
I work across Rust, Python, and TypeScript, from knowledge models and data pipelines to APIs and web interfaces.
Evidence before fluency. Time as first-class data. Human review over silent automation. Portable context over disposable prompts. Public repositories distinguish demonstrated behavior from planned work and state their limitations directly.
At Concordia Discors, I write about international affairs, institutions, pluralism, and the human questions behind technology.
- On minilateralism and institutions that work
- The conflict resolution machine and the human remainder
For foreign-policy research, collaboration, or a conversation about these tools: connect on LinkedIn.