Alexander Towell · queelius · metafunctor.com
PhD student (SIU) · MS Math & Stats · MS CS · Stage 4 cancer. I keep working.
Reliability theory, encrypted search, algebraic structures, complex networks.
- masked-causes-in-series-systems — likelihood framework for masked failure data
- encrypted_search_thesis — IR on encrypted collections
- bernoulli_sets — algebraic theory of random approximate sets, closure properties, applications to encrypted Boolean search
- cognitive-mri-conversations — network analysis of AI conversations (Complex Networks 2025)
- algebraic.dist — algebra over distributions in R
- likelihood.model — composable likelihood framework
I want what I make and collect to be readable in a hundred years. These are Unix-philosophy CLIs for owning your own data — plain formats, composable through pipes, no dependency on services that might disappear.
| Tool | What |
|---|---|
| longecho | Self-describing, plaintext-first archival philosophy with graceful degradation. Self-similar: every source in a longecho is itself longecho, so site generation, search, and JSON export apply recursively. |
| arkiv | Universal data archive (longecho compliant) |
| repoindex | Git repo collection index |
| btk | Bookmarks |
| memex | All my AI conversations, searchable and queryable |
| mtk | Email archives |
| ptk | Photos |
| ebk | Ebooks |
| chartfold | Personal health data from multiple EHR systems |
| deets | Personal metadata |
| jot | Notes & tasks |
| crier | Cross-posting |
| jsonl-algebra | Relational algebra over JSONL |
| posthumous | Deadman switch |
| eidola | Conversable persona from all of the above |
chartfold consolidates my health records from multiple hospital systems into one place. I do a lot of analysis on my own data — trying to spot patterns, understand trends, inform conversations with my doctors. They're the experts; this just helps me be a better-informed patient.
eidola uses arkiv and longecho to auto-package all my data — conversations, writings, emails, bookmarks, photos, reading notes — into a queryable whole (including MCP/SQL). The goal is a simulacrum: something that can represent my thinking after I can't. The cognitive MRI paper is the academic version of the same impulse — mapping the topology of a mind's conversations.
My thesis was about searching encrypted data without leaking queries. That research turned practical — I keep building tools for the same problem: keeping things private on infrastructure you don't control.
- cryptoid — Encrypted content and user management for Hugo/static sites
- pagevault — Password-protect HTML, Markdown, PDFs, or any content on static hosting. User management, re-keying, renders to a single-file app.
- bernoulli_sets — Theory of random approximate sets underlying encrypted Boolean search
- encrypted_search_confidentiality — Measuring and maximizing confidentiality in encrypted search
- boolean-algebra-over-trapdoor-sets — Boolean algebra on trapdoor sets for oblivious computing
I build a lot of tooling for working with AI — Claude Code plugins, MCP servers, agent workflows.
- alex-claude-plugins — Plugin suite: papermill (academic writing), pub-pipeline (publication), worldsmith (fiction), repoindex, mf (site management), deets, locksmith
- eidola — Persona generation from personal data archives
- cognitive-mri-conversations — Network topology of AI conversations
C++: disjoint_interval_set, algebraic_hashing, sparse_spatial_hash, elementa, limes
R: algebraic.dist, hypothesize (CRAN), likelihood.model, nabla
Python: AlgoTree, rerum, fuzzy-infer
- The Policy — literary SF: AI alignment, consciousness, emergence (published)
- Echoes of the Sublime — philosophical horror: AI safety, cognitive bandwidth (~103k words)
ORCID · Scholar · PyPI · Blog · [email protected]