Disability-led research software: telemetry research, C++/HLS prototypes, RAG guardrails, and offline evidence-audit tools, with scope and evidence status documented per project.

sparkainlpx.xyz · LinkedIn · Research notebook · Live demo: Holographic Thoughts · ORCID · Zenodo community
Français (langue maternelle) · English (fluent)

Open to work: applied AI research and research-software roles, remote within Canada or on site in Fredericton. Contact details are below.

Spark AI NLP is a disability-led research-software company based in Fredericton, New Brunswick, Canada. The public repositories are research prototypes: a reproducible telemetry anomaly-detection benchmark, the OES-32/OES-512 residual-latch family, C++/HLS kernels, a source-grounded RAG guardrail, two offline audit tools (Evidence Passport for run-level evidence, Quantum Claims Evidence Passport for public claims against their sources), and small offline tools for provenance and consent-first data sharing. They are not hardware products, field products, or medical products, and they do not report results on quantum hardware.

Six repositories that show the range of the work. Each README states its own scope, limits and evidence status.

Every public repository belongs to one of six tracks. Repositories stay separate so that each keeps its own licence and release history, and its own Zenodo DOI once released.

Repositories not named here are listed in the sections below and on the repository list.

Each project is placed on one rung. The rung says what exists today, not what is hoped for.

The rungs sit on top of the evidence tags used in every repository: SYNTHETIC (generated inputs), REPORTED (measured, with the source and protocol named), TARGET (a design goal) and UNRUN (scripts exist, the run has not happened).

• Everything here is classical software. Results are SYNTHETIC unless a README names a measured dataset and its protocol, and then they are tagged REPORTED.
• No repository reports a result from quantum hardware, a physical membrane, a medical device or a field deployment.
• Nothing is described as established beyond what its tests and data support. Negative results are published as measured.
• Unrun work (hardware synthesis, hardware-in-the-loop, independent review) is labelled UNRUN, and design goals are labelled TARGET.
• If a statement here or in a repository looks unsupported, open a Claim correction issue; the template is in sparkainlp-x/.github.

• OES-61440 tripwire (in identity-runtime): 61,440 fixed threshold channels, grouped as 1,920 membranes of 32 channels (120 sectors × 4 supervisors × 4 membranes = 1,920; 1,920 × 32 = 61,440). Each channel compares a fixed weighted sum of four risk signals with a fixed threshold.
• Nothing learned: weights and thresholds come from SHA-256 of the channel index. The settings are illustrative, not calibrated against incident data, and the tripwire is not a safety guarantee; a plain rule holds any action with effective risk of 0.5 or more for a human.
• Hash-chained ledger: every step is appended to a JSONL ledger in which each entry's hash covers the previous entry's hash. Editing, reordering or deleting lines breaks verification. The ledger is tamper-evident, not tamper-proof.

• C++ software and HLS automation scripts for a qLDPC decoder targeting the AMD ZCU111 RFSoC; see qldpc_decoder_cpp.
• CI builds and tests the portable C++ code. FPGA synthesis, timing closure and hardware-in-the-loop runs are UNRUN, and no qubits or quantum hardware are involved. There are no hardware results.

• Deterministic residual-telemetry triage across the OES-32/OES-512 family: fixed 32-channel blocks, transparent scores and fail-closed input contracts, with one normative residual definition (oes32-residual, ADR-001).
• Detectors are benchmarked against simple baselines and the outcomes are published as measured, including negative results; see Featured work.

This index links project names to their canonical public repositories. A direct match identifies the corresponding public project; it is not a claim of scientific validation. Related projects are separate implementations or tools, and their individual READMEs define their scope and evidence.

The profile and repository READMEs distinguish synthetic or theoretical results from measured results, targets, and runs that have not been performed. Code quality or reproducibility alone does not establish scientific validity.

• membrane-listen · v0.1.0: SYNTHETIC sonification of a classical phase-oscillator ring (standard-library Python, AGPL-3.0-only); a pre-registered study with confidence intervals in which 2 of 23 hypotheses were not supported; not a recording, not a medical model; no listening test was run. DOI 10.5281/zenodo.23687758
• membrane-heart · v0.1.0: SYNTHETIC toy in which a heart-shaped network of 64 simulated neurons carries a phase wave stored in an associative memory with a calibrated gate that names or refuses a probe (standard-library Python, AGPL-3.0-only); not a medical model, not a diagnosis. DOI 10.5281/zenodo.23680672
• identity-runtime: standard-library Python runtime for governed, reversible avatar personalities, with human approval quorums, expiring capability leases, the OES-61440 tripwire and a hash-chained ledger; 99 tests in CI; not tested against real attackers; demo page. DOI 10.5281/zenodo.23448946
• infinite-becoming: browser-playable synthetic ecosystem game on a reaction–diffusion substrate; uncalibrated; its frozen persistence regression is disclosed (1 of 5 seeds); play online. DOI 10.5281/zenodo.23423943
• world-holobridge: standard-library Python + SQLite reference prototype of a governance workflow for community-led cultural and spatial-media projects, with a bilingual browser plan generator; not legal advice; no community endorsement claimed (MIT OR Apache-2.0); plan generator. DOI 10.5281/zenodo.23423941
• spark-evidence-lab: rerunnable experiment cards that put my methods next to simple baselines, including where the baseline wins; every number copied from a committed source file; live page. DOI 10.5281/zenodo.23403169
• digital-to-wave-2-thz-validity: engineering-validity study of the THz equations in the Digital to Wave Part II preprint (equation inventory, dimensional checks, protocol-locked synthetic channel benchmark); SYNTHETIC; no physical, biological or safety claim. DOI 10.5281/zenodo.23375690
• orbit-wave-oes-lab: software-only path from a stylized orbit timeline to a synthetic signed-wave payload, fixed-threshold readout, optional OES replay and evidence passport, plus offline binaural capture analysis; SYNTHETIC; orbit, spacecraft, quantum and acoustic hardware UNRUN. DOI 10.5281/zenodo.23372536
• spark-membrane: fail-closed console that audits SYNTHETIC 32-channel frames against the OES repositories, pinned by commit SHA, with a hash-chained run trail, evidence passport and claims gate; thresholds uncalibrated; demo page. DOI 10.5281/zenodo.23175490
• oes-telemetry-bench: offline replay bench for synchronized 32-channel OES32 telemetry frames, with a strict input contract and frozen thresholds; synthetic demo. DOI 10.5281/zenodo.23175492
• residual-gradient-workbench: browser workbench that keeps an OES-32 residual replay and a classical thermoelectric generator model in separate lanes; demo page. DOI 10.5281/zenodo.23187280
• thermoelectric-ledger: signed heat-flow accounting that withholds exergy and efficiency figures unless the uncertainty bounds support them; 104 contract tests in CI (MIT). DOI 10.5281/zenodo.23241554
• qec-scaling-bench: classical toy simulation of surface- and repetition-code memory scaling with Stim and PyMatching; seeded and reproducible, no hardware results. DOI 10.5281/zenodo.23241631
• flip-observer-1024: calibration prototype for a classical weighted readout under a hidden channel-order reversal (512 inputs, two noisy banks); SYNTHETIC; demo page. DOI 10.5281/zenodo.23173449
• oes-quantum-binaural-lab: classical toy simulation with three separate lanes (Bell-correlation Monte Carlo, scalar metrics ledger, digital stereo loopback); SYNTHETIC. DOI 10.5281/zenodo.23175494
• signal-loom: offline performance-art prototype in which synthetic signal streams drive an interactive visual score (MIT); demo page. DOI 10.5281/zenodo.23173453

64 of the 75 public repositories have Zenodo DOIs (count as of 2026-10-24); the newest repositories are being released. Use each repository's README for its citation and DOI details. Records are being added to the Spark AI NLP: Research Software Zenodo community.

• Minimal dependencies. Many tools use only the Python standard library (for example identity-runtime, world-holobridge, evidence-passport and multi-quantum-oes), and infinite-becoming has no third-party runtime dependencies. Others depend on what the work needs: oes-resilience requires NumPy, qec-scaling-bench uses Stim and PyMatching, and qldpc_decoder_cpp builds against the upstream ldpc library.
• Tests and CI per repository. Test suites and GitHub Actions workflows are maintained repository by repository; identity-runtime and oes-resilience, for example, run in CI on Python 3.10–3.13.
• Open licences, per repository. Most repositories are AGPL-3.0-only with a commercial-licence option; some are MIT, and world-holobridge is MIT OR Apache-2.0. Each repository's LICENSE file is authoritative.
• Archived releases. Released versions are archived on Zenodo with concept and version DOIs (see Cite).

• Contracts first. Each definition lives in one normative place with executable tests. For OES-32 that is oes32-residual@b77b612 (ADR-001); other implementations are labelled Profile A sidecars and document how they differ.

• Fail-closed where specified. Selected READMEs label unrun or unmeasured work; results and caveats are documented per project, including negative outcomes where evaluations were performed.

• Evidence labels where reported. Project READMEs define their own evidence status; labels and coverage are not universal across every repository.

  Tag	Meaning
  SYNTHETIC	Produced by our own simulation or tests on generated inputs
  REPORTED	A project result based on identified measurements or benchmark data. The README states whether data were collected for this project or pre-existing, names the source and conditions/protocol, and says whether independently reproduced
  TARGET	A design goal; not yet achieved or measured
  UNRUN	Tooling or scripts exist; the run has not been performed

• Reproducibility varies by project. Some repositories document tested Quickstarts, automated tests, or CI; check the individual README and workflow for current coverage.

• Accessibility first. Ask for alternative formats by opening an issue in any repository.

Contribution guidelines, the code of conduct, and the security policy are shared across all repositories: sparkainlp-x/.github.

64 of the 75 public repositories have Zenodo concept and version DOIs (count as of 2026-10-24), and each released repository includes CITATION.cff. Use GitHub's "Cite this repository" button or the citation links in the specific repository when available. Records are being added to the Spark AI NLP: Research Software Zenodo community.

Mun Seok Lee and Somnath Laha are acknowledged as collaborators. Lee's Digital to Wave preprints and packages received under his name are the source for the Digital-to-Wave repositories and the membrane-related repositories that say so in their Provenance sections. Laha's related discussion of a satellite concept motivated the framing of orbit-wave-oes-lab. This is an acknowledgement only: it implies no endorsement of these repositories and no authorship. The software is by Jean-François Brisson, and each repository's README states what it derives from.

Open to applied AI research and research-software roles (remote within Canada, or Fredericton, NB), and to collaboration with research groups, FPGA/decoder engineers, and accessibility-focused partners.

• Website: sparkainlpx.xyz
• LinkedIn: Jean-François Brisson
• Research notebook: Spark AI NLP notebook
• Live demo: Holographic Thoughts (interactive toy views; synthetic, classical simulation)
• ORCID: 0009-0000-9778-5374
• Questions about code: open an issue in the relevant repository