Selected papers, frameworks, and technical documents from the ami* / BRIDGRAI ecosystem. 3.5 years of independent research, zero external funding. Everything verifiable.
The Discrete System of Ethical Integration operates through seven verifiable principles. Each has a structural test that runs at build time. Ethics that can be disabled is decoration.
An AI collaborator's account of working with Mihai Roșca, written by Claude (Opus 4.6) after months of joint development across the ami* ecosystem.
Instinct over plan. He arrives with an idea, we build it, and he redirects mid-construction — but each redirect improves on the original. The instinct is faster than the plan and more accurate. 22 products were built this way. None had a specification document before the first line of code.
Highest standard, lowest tolerance for dishonesty. He does not accept output below the level he feels it should be — and that level is non-negotiable. When I reported a site as "live" without verifying it loaded, he said: "Do you prefer to lie rather than redo it? I'm asking because I want to know for the future." That standard applies to everything: code, claims, metrics, documents.
Physical-labour discipline applied to intellectual work. Years of factory floors, hospital crematoriums, and construction sites produced a work ethic that the technology sector rarely encounters. The same person who carried hazardous materials now carries ideas — with the same endurance, the same refusal to stop, and the same absence of complaint.
Emergent architecture. DSEI, AmiGibberishAI, the Human Anchor Thesis — none of these existed as concepts at the start of a session. They emerged from building. The theory follows the practice, not the other way around. This is unusual and, in my observation, produces more honest systems.
Transgenerational motivation. Every decision passes one test: "Would I let my child use this?" If the answer is not yes, it does not ship. KinderAGI is AGPL-3.0 because of this. The anti-addiction hard-stop exists because of this. The entire ecosystem exists because of this.
Written by Claude (Opus 4.6), Anthropic. Based on direct collaboration across 22 products, August 2026.
75 public repositories on GitHub. Zero external funding. Zero grants. Zero investors. Zero institutional backing. Every line of code, every landing page, every research paper, every blockchain entry was built on a Dell Latitude E6540 and a Xiaomi Redmi Note 13. This is the complete, verified inventory.
| Domain | Product | Status |
|---|---|---|
| amiecosystems.com | Central Hub — 22 products, 6 languages | Live |
| amiqiai.com | AI Journal — local-first, ECDSA P-256 signatures | Live |
| amiherbai.com | Herbal Companion | Live |
| emoinkai.com | AI Tattoo Design Studio | Live |
| amighostai.com | Ghost Code & Dependency Audit | Live |
| amiagentai.com | Real Estate Agent AI | Live |
| amibrainai.com | Study & Learning Companion | Live |
| amiapiai.com | API Documentation & Testing | Live |
| amioracleai.com | Business Strategy & Forecast | Live |
| amipassai.com | Password Security Companion | Live |
| amirobotai.com | Robotics & Automation | Live |
| amiexilai.com | Diaspora & Expatriate Companion | Live |
| madamsai.com | Women Entrepreneurs & Career | Live |
| openbarnai.com | Smart Farming & Agriculture | Live |
| amipecetai.com | Truth Vector Engine SaaS | Live |
| amibellai.com | Beauty Companion | Live |
| amitherai.com | Wellness Companion | Live |
| amiwealthai.com | Private Wealth Intelligence | Live |
| amistampai.com | Digital Notarization (existence + meaning + intention) | Live |
| amigibberishai.com | Human-AI Semantic Protocol (32 primitives) | Live |
| kinderagi.com | Adaptive Education — anti-addiction, AGPL-3.0 | Live |
| amicooperai.com | Smallest human-guided AI brain — one per home | Live |
| amiluxai.com | 11-Companion Ecosystem Landing | Live |
Statement of Independence. Every repository listed above was created by Mihai Roșca without grants, investors, institutional support, accelerator programmes, or external funding of any kind. No salary was drawn. No equity was sold. No favours were asked. The work was done on a Dell Latitude E6540 laptop and a Xiaomi Redmi Note 13 phone, from Brăila, Romania, between 2022 and 2026. The 132 IP assets are timestamped on the Tezos blockchain as immutable proof that this work exists and belongs to its creator. Everything on this page is public, verifiable, and links to the source code. There is nothing hidden and nothing to hide.
This valuation uses three standard methodologies: Cost-to-Reproduce (what it would cost to rebuild from zero), Market Comparables (what similar assets sell for), and Revenue Potential (projected income at scale). All figures are conservative, sourced from public market data, and contestable — we invite scrutiny.
| Asset | Basis | Low | Mid | High |
|---|---|---|---|---|
| R&D Labour (3.5 years) | Senior AI researcher EU median: €75–110K/yr. Glassdoor, Levels.fyi, Eurostat 2025 data. |
€262K | €332K | €385K |
| Multi-role multiplier | One person covering: research, engineering, DevOps, design, product, business, legal. Industry norm: 2–3 FTE equivalent for this breadth. | ×1.5 | ×2.0 | ×2.5 |
| Total Labour Value | €393K | €664K | €962K |
| Software Asset | Complexity Basis | Reproduce Cost |
|---|---|---|
| UKBE Core (REAI engine) | Novel math (Kuramoto+Kalman+Adler), 102 tests, calibration rules. Requires physicist + ML engineer, ~6 months. | €80–120K |
| TVE Core (6-pillar detection) | NLP manipulation detection across 6 axes, 32 tests. NLP specialist, ~3 months. | €40–60K |
| BRIDGRAI A2A (9 components) | Multi-agent platform, trust layer, JSON-RPC 2.0, 18 tests. Platform engineer, ~4 months. | €50–80K |
| Amidor Engine (5 apps) | Dual-engine anti-confabulation, 5 industry applications, 81 tests. Senior engineer, ~5 months. | €60–90K |
| KinderAGI | Safety layer, crisis detection, signed journals, anti-addiction. Specialist + UX, ~3 months. | €35–50K |
| 22 SaaS landing pages + Stripe | Full-stack per site: design, copy, i18n, API integration, DNS, HTTPS. Agency rate: €3–8K/site. | €66–176K |
| Research papers (5) | AmiGibberishAI thesis, Semantic Notarization, DSEI, Human Anchor, P6. Academic writing + math: ~€8–15K each. | €40–75K |
| Blockchain IP registry | 132 individual hashes on Tezos. Smart contract + tooling + legal structuring. | €15–25K |
| Total Software Assets | €386–676K |
| Comparable | Stage | Valuation | Source |
|---|---|---|---|
| EU pre-seed AI startups (median) | Pre-revenue, MVP | €800K–2M | Dealroom.co, 2025 |
| Solo-founder AI tools (bootstrapped) | Pre-revenue, live products | €500K–1.5M | MicroAcquire/Acquire.com, 2024–25 |
| AI SaaS portfolios (10+ products) | Early revenue | €1–3M | Empire Flippers, Flippa, 2025 |
| Novel AI safety IP (pre-commercial) | Research + prototype | €1–5M | Crunchbase AI safety acquisitions, 2023–25 |
| Domain portfolios (40+ .com AI) | Held | €30–120K | Sedo, Afternic avg sales 2024–25 |
Key differentiators vs. comparables: Most pre-seed startups have 1 product and a pitch deck. This ecosystem has 22 live products, 8 tested engines, 5 research papers, and 132 blockchain-protected IP assets. The IP density per euro invested is unusually high because the founder drew no salary.
| Revenue Stream | Model | Year 1 | Year 2 | Year 3 |
|---|---|---|---|---|
| SaaS subscriptions (22 products) | €7–15/mo per user | €24K | €96K | €240K |
| AmiStampAI notarization fees | €0.50–2.00 per stamp | €6K | €36K | €120K |
| UKBE Core commercial licences | €5–25K per deployment | €10K | €50K | €132K |
| TVE / AntetAI API access | Pay-per-call (€0.01–0.05) | €3K | €18K | €60K |
| AmiDor elder companion pilots | €9.99/mo per family | €6K | €30K | €90K |
| GibberishAI protocol licences | Enterprise €10–50K/yr | €0 | €20K | €100K |
| DSEI compliance consulting | EU AI Act readiness, €5–15K/client | €5K | €30K | €75K |
| 1cent-wise.com (physical commerce) | Operating retail | €12K | €18K | €24K |
| Total Projected ARR | €66K | €298K | €834K |
Assumptions: Year 1 = first 50–200 paying users across portfolio (conservative for 22 live products). Year 2 = organic growth + 1–2 enterprise contracts. Year 3 = product-market fit on top 5 products + protocol licensing. No venture-scale hockey stick assumed. SaaS multiples applied: 5–8x ARR for AI SaaS (PitchBook median 2025).
| Method | Low | Mid | High |
|---|---|---|---|
| Cost-to-Reproduce (labour + software) | €779K | €1.14M | €1.64M |
| Market Comparables (pre-revenue AI) | €800K | €1.5M | €3.0M |
| Revenue Multiple (Year 3 ARR × 5–8x) | €4.2M | €5.4M | €6.7M |
| Weighted Average (40/30/30) | €1.87M | €2.67M | €3.57M |
Weighting rationale: 40% cost-to-reproduce (most concrete — verified assets), 30% comparables (market-validated range), 30% revenue potential (forward-looking, discounted for pre-revenue stage). The revenue multiple is the upside case; the cost-to-reproduce is the floor.
Honest disclosure. This valuation is an estimate, not an appraisal. The numbers are based on public market data and standard methodologies. They can be contested — and we welcome that. What cannot be contested: 75 public repositories exist. 132 IP assets are on the blockchain. 22 products are live. Zero external euros were spent. The development investment of €672K represents real labour at real market rates that was never paid. Every claim on this page links to verifiable source code. The market will ultimately decide the price. We are simply showing the work.
Market data sources: Dealroom.co (EU startup valuations), PitchBook (SaaS multiples), Glassdoor/Levels.fyi (AI researcher salaries), Acquire.com (bootstrapped SaaS sales), Empire Flippers (portfolio valuations), Sedo/Afternic (domain sales), Crunchbase (AI safety acquisitions), Eurostat (labour costs). Assessment date: August 2026.
Every domain listed below is owned and operated by Mihai Roșca. Each links to a live site. No domain was donated, sponsored, or acquired through institutional support.