Independent AI Safety Researcher · Senior QA Engineer · FinTech reliability background

I build verifiable trust infrastructure for AI agents, fintech actions, memory systems, and public-good protocols.

High-risk AI-agent actions should be reviewable, replayable, and evidence-backed before execution.

My work focuses on the infrastructure layer between an AI-agent proposal and a real-world effect: tool calls, code changes, infrastructure actions, financial workflows, governance actions, public-good protocols, and other high-impact operations.

🌐 RESONANCE — Journal of Intelligence, Technology & Human Progress

I am the creator and editor of RESONANCE, an independent, evidence-first global journal about AI agents, trust infrastructure, science, technology, startups, and human progress.

RESONANCE is the readable public layer above my research repositories: GitHub keeps the code, fixtures, benchmarks, PRs, and verification trail; the journal turns that evidence into clear stories, testable ideas, and open questions that engineers, researchers, founders, and product teams can actually engage with.

Why RESONANCE is different:

• Evidence-first: important claims link back to inspectable source material, code, PRs, fixtures, benchmarks, or authoritative references.
• Readable research: dense technical threads become clear narratives — problem → mechanism → evidence → verification → implication → action.
• Multilingual: key publication paths are available in English, Russian, and Simplified Chinese.
• Market dialogue, not vanity metrics: selected articles end with a concrete open question designed to surface real workflows, missing guarantees, and product needs.
• Research-to-product loop: PUBLISH → ASK → LISTEN → DIAGNOSE → PILOT → PROVE → BUILD.
• Inspectable publication quality: SEO, internal-link, and Lighthouse experience contracts run in CI before publication.

Recent research: Memory Can Be True and Still Be Unsafe — why historical truth should not automatically retain current authority to drive an AI-agent action.

Open the journal → RESONANCE
Source repository → safal207/RESONANCE

NVIDIA — Open Secure AI Alliance inquiry submitted (29 July 2026). I submitted the official alliance form as an independent open-source maintainer and sent follow-up updates to the NVIDIA Connect and NVIDIA Inception program teams about the open-source agent-assurance portfolio.

Status: awaiting guidance on the appropriate contributor or ecosystem pathway.

This is a factual status update only. It does not imply membership, acceptance, affiliation, partnership, endorsement, or adoption by NVIDIA or the alliance.

My post-sandbox agent-governance work is being shared for adversarial review. The goal is to collect counterexamples, reproducible failure traces, and evidence-backed review state — not logos or implied endorsements.

Important: SENT, ACKNOWLEDGED, and ROUTED are review-process states. They do not mean endorsement, independent reproduction, adoption, or validation.

The review trail is itself machine-verifiable:

external response
      ↓
Review Event Envelope
      ↓
evidence-backed state transition
      ↓
External Validation Graph / EEW
      ↓
Git + CI
      ↓
ProofPath / CML

Evidence source: LiminalOSAI — post-sandbox governance review trail

Aleksey Safonov (Alex Lim, @safal207) — founder and evidence-systems builder

QA engineer, AI-product builder, and entrepreneur focused on turning uncertain, high-stakes ideas into testable protocols, traceable artifacts, and bounded decisions.

Mission: build trustworthy infrastructure where AI models may propose possibilities, but evidence contracts and authorized humans determine what is supported and what may proceed.

Working principles: root causes over symptoms · evidence over confidence · power under control · one verifiable artifact at a time.

Current frontier: Kairos Gate for X-Cell, an evidence and governance layer for agentic biology that separates scientific support from permission to act.

Role boundary: I do not present myself as a biologist or clinician and do not authorize experiments or treatment. My contribution is QA, product thinking, causal and evidence architecture, reproducibility, traceability, and governance.

Personal archetype: the Silver Surfer — great power restrained by principle, balance, and responsibility.

🧬 Kairos Gate for X-Cell

Evidence and governance infrastructure for agentic biology.

Kairos Gate checks experimental units, provenance, independent replication, temporal identity, competing causal explanations, model compatibility, and claim boundaries before a scientific conclusion is allowed to move forward.

models may propose possibilities
                ↓
evidence contracts determine what is supported
                ↓
authorized humans determine what may proceed

The project is computational and documentary only. It does not authorize wet-lab work, treatment, or clinical decisions.

Open the bio direction → Kairos Gate for X-Cell

Kairos Gate is being designed to audit, compare, or interoperate with public open-source biology projects. These are technical and scientific ecosystem relationships, not claims of affiliation, endorsement, integration, or formal partnership.

public model or dataset
          ↓
exact-version provenance
          ↓
Kairos evidence contracts
          ↓
supported, limited, unresolved, or blocked claim

A repository appearing in this map means it is a relevant public evidence, model, benchmark, or agent surface. It does not mean its maintainers have reviewed, approved, adopted, or partnered with Kairos Gate.

Modern AI systems do not only answer questions. They increasingly call tools, write code, modify infrastructure, move data, trigger workflows, and make decisions with real-world consequences.

My research and engineering work asks:

• What evidence should exist before an AI agent performs a high-risk action?
• How can action traces be made replayable, tamper-checkable, and useful for human reviewers?
• How can deterministic control layers complement probabilistic model evaluations?
• How can financial, governance, and infrastructure actions become more accountable before execution?
• What should a practical trust layer look like for multi-agent and human-in-the-loop systems?

Independent evidence-first journal and market-dialogue layer for AI, trust infrastructure, science, technology, startups, and human progress.

research signal -> evidence -> readable story -> open question -> real workflow -> product signal

RESONANCE makes the technical work easier to inspect and easier to discuss outside a repository. It connects public research artifacts to a multilingual editorial layer and a structured PUBLISH → ASK → LISTEN → BUILD loop.

Evidence and governance layer for agentic biology.

scientific proposal -> evidence audit -> bounded claim -> authorized human decision

Kairos Gate separates biological evidence, model output, and permission to act. It is the main bio-direction project for experimental-unit auditing, provenance, replication, temporal validity, causal-hypothesis ranking, and partner-laboratory evidence handoff.

Causal audit layer for AI-agent action traces.

agent action -> parent cause -> audit rule -> finding -> reviewer evidence

CML focuses on causal lineage: whether an agent action has a valid parent, approval path, and reviewable trace. It is designed as a lightweight accountability primitive for agent frameworks, memory systems, and high-risk tool-use workflows.

Pre-execution gateway for high-risk AI-agent/API actions.

valid credential != valid action != valid scope != valid reversibility != valid approval

ProofPath protects the meaning of an action before execution: intent, scope, causal parent, reversibility, approval, and audit trail.

Best entry point for reviewers:
Reviewer First Screen

Local-first and federated event-memory database for inspectable, replayable systems.

LiminalDB explores event envelopes, local replay, validation paths, and future federated replication for memory and protocol infrastructure.

Public-good test kit for Fediverse portability, export/import checks, media integrity, visibility safety, and reviewer-friendly compatibility reports.

Research and portfolio space for pre-execution evidence gates and AI-agent oversight prototypes.

Best entry point for grant, fellowship, and AI safety reviewers:
Reviewer Start Here

Causality-aware QA/CI reliability substrate for reproducible failure analysis and quality decision packets.

Best entry point for reliability and open-source infrastructure reviewers:
Reviewer First Screen

The broader project family is becoming a layered stack:

CML                         -> causal accountability for agent traces
ProofPath                   -> action authorization and payment/API guard
LiminalDB                   -> replayable event-memory substrate
L-THREAD / LTP              -> secure trace and replay protocol
Liminal Presence Interface  -> presence, identity, and interaction layer
L-EDGE                      -> edge/runtime direction
Fediverse portability kit   -> public-good validation and portability testing

This is the long-term direction:

verifiable trust infrastructure for AI agents, fintech actions, memory systems, and public-good protocols

These projects are related to the long-term platform direction, but should remain secondary until the active focus is clearer:

• L-THREAD-Liminal-Thread-Secure-Protocol-LTP-
• Liminal-Presence-Interface-LPI
• L-EDGE-Liminal-Edge-OS
• DMP-decision-memory-protocol
• Living-Relational-Identity-LRI
• LRE-Core
• Access-Orientation-Protocol
• Scale-Descent-Protocol-SDP

These projects connect my background in QA, fintech, reliability, and product systems:

• CareerOS
• qa-fintech-api-python-course
• finanalytics-core
• TachTachAI
• nexus-sales
• nexus-ecosystem

Some repositories are intentionally kept as prototypes, sketches, or idea bank material. They may contain useful concepts, but they are not the current execution focus.

Examples include older experiments around Liminal, Noosphere, education, Web3, DAO, voice, self-creation, and personal protocol ideas.

These projects are experimental open-source research prototypes, not production safety infrastructure yet.

They do not claim full AI alignment, complete agent safety, certified security, regulatory compliance, or universal prevention of unsafe actions.

The current contribution is narrower and more testable:

make high-risk AI-agent actions reviewable, replayable, and evidence-backed before execution

I have 12+ years of software QA and FinTech reliability experience, including brokerage, banking, API, WebSocket, SQL, risk, reporting, test strategy, regression prioritization, and quality process design.

This background shapes my AI safety work: I treat agent oversight as an engineering reliability problem, not only as a model-behavior problem.

For reviewers, grantmakers, and collaborators:

• RESONANCE — evidence-first journal
• RESONANCE source repository
• Kairos Gate for X-Cell — bio direction
• Causal-Memory-Layer
• ProofPath reviewer first screen
• PythiaLabs reviewer start
• PythiaLabs portfolio map
• LiminalQAengineer reviewer first screen
• LiminalDB
• Fediverse portability test kit

Email: safal0645@gmail.com
Telegram: @Alexfox14
GitHub: https://github.com/safal207

I build deterministic oversight layers that gate, audit, and explain high-risk AI-agent actions before execution — and publish the evidence and open questions through RESONANCE.