Independent researcher & developer building rigorous, auditable systems for scientific computing, mathematical modeling, AI reasoning, and verification.
Reproducible computation · explicit assumptions · deterministic verification · falsifiable models
Independent developer and researcher working across AI systems and scientific computing, mostly on software where the reasoning is explicit rather than implicit: numerical algorithms you can read end to end, models whose assumptions are written down, and AI agents whose claims are checked by deterministic machinery instead of trust.
The work runs along two threads:
- Scientific computing from first principles — from-scratch numerical libraries with zero runtime dependencies, where the algorithm is the documentation, alongside a NumPy/SciPy production line for accelerated and optionally GPU-backed work.
- Deterministic verification for AI — turning model output and agent plans into sealed contracts, hash-chained evidence, and PASS / FAIL gates, so that done means here is the command output that proves it.
Everything is built openly and tested: property, fuzz, oracle, cross-platform, and benchmark-regression suites run in CI, with provenance attestations carried on published artifacts where the packaging supports it.
| Project | ⭐ | Stack | What it does |
|---|---|---|---|
| polyglot-python |  |
Python · MIT | A deterministic keyword transpiler: write Python with localized keywords in eight languages — Turkish, Spanish, Portuguese, German, French, Chinese, Hindi, Arabic — and it maps them to real Python and hands the result to CPython. No AI, no network, no randomness. Not a prose-to-code translator. PyPI + npm wrapper; Python 3.12+. |
| scientific-computing-system |  |
Python · MIT | Pure-Python computational science platform with zero runtime dependencies: linear algebra, ODE/PDE solvers, Monte Carlo, statistics, signal processing, uncertainty and dimensional analysis, reproducible workflows. Published on PyPI; npm ships a Python CLI launcher, not a JavaScript engine. Python 3.11+. |
| axiomize |  |
Python · MIT | The scientific modeling layer (the numerical library lives in scientific-computing-system): explicit, versioned Model IR, mandatory units, dimensional/numerical validation, calibration, uncertainty analysis, SBML/CellML/Modelica export. CLI, REST and MCP (axiomize mcp); pip install axiomize. |
| scientific-computing-system-2.0 |  |
Python · MIT | The NumPy build of scientific-computing-system — not a separate product: the NumPy/SciPy/pandas/matplotlib line with guided fitting, held-out diagnostics, replay manifests, drift checks, optional compiled C kernels and CuPy GPU execution. PyPI + npm. |
| plan-auditor |  |
Python · MIT | Makes an agent's done mean here is the command output that proves it: LLM-free verification with sealed plans, SHA-256 hash-chained evidence, STRIPS/PDDL reachability and a fail-closed PASS / FAIL / UNKNOWN gate. PyPI + npm. |
| quantum-reasoning-skill |  |
Python · MIT | Model-agnostic Agent Skill that keeps genuinely different hypotheses alive, tests each against evidence, revives alternatives and collapses to the best-supported answer. Quantum-inspired metaphor, not quantum computing. Status: protocol and controller benchmark published; model evaluation pending. PyPI + npm. |
| eq-layer |  |
Python · MIT | A control layer that decides how an LLM should speak before it speaks: affect/intent tracking, Bayesian-risk action routing, conversation repair, policy steering. Early stage; evaluations preliminary; not a factual truth checker. |
| nomosguard |  |
Python · Apache-2.0 | Deterministic security reasoning core: LLMs produce evidence; deterministic machinery decides. Append-only hash-chained evidence ledger, rule-derived threat/access paths, fail-closed deny-by-default policy gate. Models never cross the decision boundary. |
| literature-evidence-mcp |  |
Python · TypeScript · Rust · Go · C++ · Apache-2.0 | Scientific literature and evidence-provenance MCP server: multi-provider search, citation and retraction checks, quote verification, evidence logs. v0.5.1 is a bibliographic/traceability and evidence-audit toolset — not an automated peer reviewer, meta-analysis engine or scientific truth oracle. |
| scientific-toolkit |  |
Python · Apache-2.0 | Provider-neutral MCP gateway for nine existing public repositories; upstream packages stay independent with their own names and versions. Published on PyPI as scientific-toolkit-mcp; local stdio and Streamable HTTP transports. An integration hub, not a source merge. |
| axiomize-quantum-skills-2.0 |  |
Python · MIT | Ships axiomize and quantum-reasoning-skill together in one package line (1.2.5) — not a separate product and not a second MCP registration. Install this one for both; install the upstream repos when you need only one. |
REPO-RELATIONSHIPS.md draws the spine, says which project to reach for,
and separates what is canonical from what is an extension, a gateway, or an early-stage experiment.
All 22 merged pull requests in repositories outside this account, highest-starred first:
| Repository | Stars | Contribution |
|---|---|---|
coollabsio/coolify |
63.3k | #12066 — fix(api): respect service_name query param in application logs endpoint |
pingdotgg/t3code |
29.0k | #16838 — fix(server): a directory no longer resolves as the resource monitor binary |
TabularisDB/tabularis |
5.2k | #1076 — feat(command-palette): add a toggle editor word wrap command (closes #1036) |
libredb/libredb-studio |
1.2k | #1637 — docs: add missing libSQL row to test-database table |
Mudlet/Mudlet |
919 | #10780 — fix: keep Lua autocomplete popup from stealing editor focus |
HelpCode-ai/anythingmcp |
773 | #1177 — feat: show build version and git commit in UI and startup log |
vaadin/web-components |
583 | #13064 — docs: add online IDE reproduction link to bug report template |
Dokploy/templates |
231 | #1171 — feat: add VersityGW template |
issuerd/issuerd |
65 | #18 — test: add proptest suites for issuerd-protocol parsers |
TheVicky1/Pact_OS |
33 | #554 — feat(ui): add animated loading state to data export button |
Fristail27/vocab-bloom-hub |
14 | #598 — i18n(pt): review the site interface texts |
nedpals/pwa-karaoke |
13 | #60 — docs: add code of conduct and issue/PR templates |
Movalabs-crew/mova-store |
10 | #408 — Fix/stellar browser uint8array |
HafidIdrissi/Time-Tracker |
7 | #367 — docs: add portable upgrade walkthrough |
crazy-goat/xdbg |
3 | #124 — fix: fall back to len(children) on malformed numchildren |
Finntegrate/tapio |
2 | #214 — docs: define a status convention for specifications |
hoslift/ranu.js |
2 | #215 — docs: add bash and zsh autocompletion recipe |
Sall25/folio |
2 | #256 — fix(ui): keep formula editor within the viewport on small screens |
GDGCampusAEC/gdgc-aec-website |
1 | #49 — perf(images): replace raw <img> with next/image across the site |
protocorn/Clippy-vision_RAN_BY_GROK_BOT |
1 | #9 — docs: add MCP setup guides for Claude Desktop and LM Studio |
ktrubilo9/edge-proxy |
0 | #65 — test: cover public and operational HTTP endpoints |
zhengkaics/opencode |
0 | #1 — fix(app): ignore none-only reasoning variants in tooltip |