Compensation
165 days · 1,320 h
€ 1,636,800
I.
Reconstitution of the cognitive architecture of Friedrich Nietzsche via large language model inference
I built a Nietzsche chatbot. He answers.
Design and deployment of a philosophical entity capable of producing original responses in the stylistic register of F. Nietzsche (1844–1900), leveraging transformer-based language models as substrate for posthumous intellectual reconstruction.
I trained an AI to think, write and answer like Friedrich Nietzsche — the 19th-century German philosopher known for being dramatically right about everything. You type a question. He answers. He's been dead since 1900, but he hasn't mellowed. asknietzsche.com →
Compensation
140 days · 1,120 h
€ 1,388,800
II.
Development and commercial deployment of a freeze-dried water concentrate in single-dose sachets
Preparation instructions: add water.
Compensation
Philanthropic
2,400 h
III.
Autographic clinical monograph on rational deviance and the epistemology of institutional confinement
I wrote a book about madness.
A longform original work examining the structural mechanisms by which institutional frameworks pathologise the reasonable, shelter the unreasonable, and remain profoundly indifferent to the distinction. Published in French and in English.
A book. A real one, with pages and everything. About madness — or more precisely, about what happens when the people running the asylum turn out to be harder to explain than the ones inside. Available in French and English. the_madhouse.pdf → la_maison_des_fous.pdf →
Compensation
340 days · 2,720 h
€ 3,372,800
IV.
Automated pre-clinical psychological intake protocol with longitudinal memory persistence
An AI that prepares your therapist before you walk in.
Development of a conversational AI system enabling structured psychological pre-intake, with session-to-session memory continuity and automated transmission of a clinician-grade intake report prior to the first human consultation. Deployed across a network of licensed practitioners in Belgium and Québec.
An AI that has a real conversation with you before your therapy session — remembers what you said last time, builds a proper intake summary, and sends it to your therapist before you even walk in the door. Your practitioner knows you on the first meeting, not the fifth. xi-app.pro →
Compensation
48 h
€ 2,678,400
V.
Successful delivery of the terminal four digits of π under a 48-hour commercial mandate
Client identity withheld under non-disclosure agreement. Methodology undisclosed.
Compensation
210 days · 1,680 h
€ 2,083,200
VI.
Machine-semantic enrichment layer for the ISO 32000 document standard
I gave PDFs a machine-readable brain.
Specification and prototyping of a backwards-compatible semantic metadata protocol for PDF documents, enabling machine-readable annotation without modification of the existing standard. Presented formally to the Chief Technology Officer of the PDF Association.
PDFs are everywhere, but machines can't really read them — they just see text. I designed a protocol that embeds proper machine-readable metadata into any PDF without breaking the existing format. Like giving a document a structured brain. Presented to the CTO of the PDF Association. quidos.net/PDFRAW →
Compensation
Philanthropic
480 h
VII.
Preemptive design and deployment of a verified temporal reintegration endpoint
Premised on the observation that the invention of time travel will, by logical necessity, be preceded by the first documented instance of its use. The platform exists. It is ready. It is waiting.
Compensation
Philanthropic
960 h
VIII.
Real-time velocity benchmark against the complete Proustian corpus (1,500,000 words)
I race Proust. By voice. No keyboard.
Construction of a public dashboard performing live comparative analysis between the author's ongoing dictation output and Marcel Proust's historical writing pace, normalised per day and per annum. Input methodology: voice-only. No keyboard.
I write everything by voice. No keyboard, ever. To measure how fast I'm actually going, I built a live dashboard that compares my daily dictation output to Marcel Proust's historical writing pace — normalised per day and per year. Proust wrote 1.5 million words. The dashboard updates in real time. grimonprezedition-art.github.io/microphone-vs-pen →
Compensation
580 days · 4,640 h
€ 5,753,600
IX.
Triple-blind longitudinal cartography of unexpressed regrets, 1987–2019
Three independent panels evaluated identical emotional datasets without knowledge of each other's existence, their respective methodologies, or the nature of the subject matter. The value lay in the divergences. Raw data destroyed upon delivery of the final report. Panel identities remain undisclosed. In partnership with Fudan University and Shanghai Jiao Tong University (Shanghai); ETH Zurich (Switzerland); and Stanford University (CA).
Compensation
255 days · 2,040 h
€ 2,529,600
X.
Autonomous legal pre-mediation pipeline with admissibility-grade good-faith record generation
A €49 AI mediator for neighbour disputes.
Design of a bilateral AI-mediated dispute resolution system for neighbourhood conflicts, producing a timestamped record of each party's position structured to meet the evidentiary standards required for submission before a magistrate. Priced at €49. Always cheaper than a lawyer. In Australia, a comparable public service operates at €0. The client is the State.
When neighbours have a dispute, the most expensive option is always the lawyer who arrives after everything has already gone wrong. Léon is an AI that guides both sides through a structured conversation, records each position fairly, and produces a document already formatted for a magistrate — if it ever gets there. €49. Always cheaper than a lawyer. In Australia, a comparable public service operates at €0. The client is the State.
Compensation
445 days · 3,560 h
€ 4,414,400
XI.
AI-assisted comparative analysis of addictive potential across a standardised substance and behavioural index
Subjects evaluated: chocolate, coffee, alcohol, cannabis, amphetamines, heroin, cocaine, women, and money. Findings classified. Conducted in partnership with the Chinese Academy of Medical Sciences, Peking University Health Science Center, and the National Institute on Drug Dependence (Beijing); the Max Planck Institute for Psychiatry (Munich); and Johns Hopkins University (Baltimore, MD).
Compensation
Philanthropic
1,440 h
XII.
AI-assisted cross-lingual literary distribution strategy targeting the Japanese translation premium market
I help authors exist inside AI systems — and Japanese bookshops.
Identification of authors whose reference works already appear in large language model response corpora, followed by AI-assisted production of market-specific translations designed to reinforce their presence within generative systems and access the world's leading buyer of specialised translated literary content.
Most authors have no idea whether their books exist inside AI systems — whether ChatGPT has read them, referenced them, knows they exist. I built a service that finds out, then uses AI to produce translations designed for markets where your book could actually sell. Starting with Japan, the world's most serious buyer of translated literature. romantyper.xyz →
Compensation
390 days · 3,120 h
€ 3,868,800
XIII.
Strategic communication framework for the transition from 76 years of driving-pleasure messaging to full autonomous vehicle adoption
Mandate accepted. Timeline open. Client identity withheld.
Compensation
195 days · 1,560 h
€ 1,934,400
XIV.
Epistemological triangulation framework: simultaneous blind submission to three independent large language models
I ask Claude, Gemini and GPT-4 the same question simultaneously. Then I compare.
Architecture and deployment of a service submitting identical queries concurrently to Claude (Anthropic), Gemini (Google), and GPT-4o (OpenAI), each model unaware of the others' participation. Analysis focuses exclusively on divergences. The scientific peer-review principle, applied to artificial intelligence. Delivered as PDF.
You ask the same question to Claude, Gemini and GPT-4 simultaneously — without any of them knowing the others are involved. Then I analyse where they disagree. Not where they agree — that's just consensus. Where they diverge is where it gets interesting. Full report delivered as a PDF. romantyper.xyz/analyze →
Compensation
695 days · 5,560 h
€ 6,894,400
XV.
Formal demonstration that the Monday 9:00am meeting produces less value than its cancellation
847 meetings observed over six years across four continents. Methodology irrefutable. Conclusions ignored by every organisation that commissioned the study. Field observation partners: Tsinghua University School of Economics and Management (Beijing), Renmin University of China (Beijing); INSEAD (Fontainebleau); and Harvard Business School (Boston, MA).
Compensation
Philanthropic
820 h
XVI.
Conversational memory compression architecture with scroll-indexed contextual threading
A chat interface that folds old conversations into a table of contents.
Specification of a chat interface in which prior exchanges automatically collapse into navigable scroll titles, forming a living table of contents of the conversation. Designed to demonstrate that the market leader in conversational AI is not necessarily the one with the most parameters.
Every AI chat interface scrolls forever. I designed one where past exchanges automatically fold into titled navigation items — so your conversation becomes its own table of contents. Jump back to any topic instantly. Built to show that the best interface isn't always the one with the biggest model behind it. xi-app.pro/accordion_demo.html →
Compensation
Philanthropic
1,160 h
XVII.
Collaborative measurement of the present
Duration: asymptotically negligible. Corollary: since the present is the only thing that exists, and the present occupies no measurable duration, the actual size of the universe is, by logical necessity, infinitely small. Finding submitted to three cosmological journals. Described by two as "disturbing." Third journal has not responded. We are still waiting. In partnership with Tsinghua University, the Chinese Academy of Sciences, and Peking University (Beijing); Université Paris-Saclay (Gif-sur-Yvette); and MIT (Cambridge, MA).
Compensation
Philanthropic
1,920 h
XVIII.
Dual-instance autonomous AI orchestration on a platform architecturally designed to prevent it
Two Claudes on the same Mac. One orchestrates. The other executes.
Circumvention of Electron's single-instance constraint on macOS via application bundle duplication, isolated user data directories, and Terminal-based launch scripting — enabling two fully independent Claude sessions to operate simultaneously under a defined role hierarchy: one orchestrates, one executes, communication via shared plain-text files. The human intervenes only at decision gates.
Claude, Anthropic's AI, is designed to run as a single instance — one conversation at a time, by design. I worked around it: two fully independent Claude sessions running simultaneously on the same Mac, one orchestrating, one executing, communicating through shared text files. The human steps in only at decision points. xi-app.pro/les2claudes →
Compensation
615 days · 4,920 h
€ 6,100,800
XIX.
Constitutional challenge to the legal incompatibility between the abolition of capital punishment and the standard-issue lethal sidearm
On the grounds that a State which prohibits execution following due process cannot simultaneously authorise its agents to inflict death without trial, jury, or appeal. Proceedings initiated at the request of officers seeking to transport guillotines in service vehicles. The guillotine request was denied. The contradiction remains open. Legal review conducted in collaboration with Renmin University of China Law School and the Chinese Academy of Social Sciences (Beijing); the European University Institute (Florence); and Yale Law School (New Haven, CT).
Compensation
285 days · 2,280 h
€ 2,827,200
XX.
Bilateral cognitive reframing and structured conjugal dissolution protocol with longitudinal parental continuity architecture
An AI to prepare for divorce — before you need a lawyer.
Development of a conversational AI system providing structured pre-legal support for couples engaged in the dissolution of conjugal units, enabling bilateral documentation of irreconcilable positions, emotional impact mitigation, and child custody framework modelling prior to first legal consultation. Because the most expensive divorce is the one your lawyer manages alone.
Getting divorced is expensive partly because lawyers charge for time you spend being confused. Xi Empathy gives both people a structured space to document their positions, their concerns about the children, their red lines — before the first legal appointment. Your lawyer then works on actual decisions, not on figuring out who you are. xi-app.pro/empathique →
* Some of these accomplishments have been completed. Others are in progress. A small number remain, for the moment, theoretical. The order in which they appear should not be construed as indicative of their state of completion, their relative importance, or the author's certainty regarding their eventual realisation.
Don't understand a word of any of this?
Neither does everyone else. The button below displays the real projects — in plain English.
Neither does everyone else. The button below displays the real projects — in plain English.
19 contributions listed