Toronto, Canada · Executive Brief · Architecture Playbook · LinkedIn · Email
I build the deterministic infrastructure around foundation models that makes them safe to run in production:
policy-gated tool execution, database-kernel tenant isolation, bounded inference economics, formal state verification, and cryptographically replayable audit trails.
Executive Brief · Flagship Systems · Control Plane Architecture · Failure Modes & Mitigations · Verified Proof Matrix · Technical Writing · About & Contact
| Engineering Pillar | How I Execute It in Practice | Shipped Public Proof |
|---|---|---|
| Architecture ↔ Implementation | Align executives, security, and product teams on system boundaries, then write the Rust, Go, C++, TypeScript, Python, SQL, and CI gates that enforce those boundaries in code. | ControlPlane (v1.0.0) · FORGE (v1.0.0) · Requiem (C++/TS) |
| Probabilistic AI, Deterministic Kernels | Treat the model as an unprivileged planner; enforce capability authorization in out-of-band MCP gateways, tenant isolation via PostgreSQL FORCE ROW LEVEL SECURITY, and untrusted execution in ZK-eBPF and Wasmtime sandboxes. |
ReadyLayer · Settler (v1.8.0) · zerokernel (v1.0.0) · WorldVM (WASM) |
| Machine-Checkable Proof Over Demos | Back architectural claims with executable specifications, Quint state-machine models, Lean 4 kernel-checked proof witnesses (Zero Axioms), Vertex AI/Gemini regression contracts, and HLO compiler delta-minimizers. |
veridag · GRAPH-Lab · UNIFORM · AgentAxiom · XLAxiom |
| Enterprise Delivery & Inference FinOps | Combine enterprise identity/compliance expertise (OAuth 2.0, OIDC, SCIM, LTI 1.3, CASE, SOC 2, FERPA) and commercial execution (McGraw Hill Solutions Architect · President's Award) with sub-millisecond token telemetry and GPU roofline modeling. |
TokenGoblin · TokPulse (v3.0.0) · Cred-bridge · Executive Brief |
< 1 ms |
FORCE RLS |
0 Syscalls |
Zero Axioms |
6 Releases |
|---|---|---|---|---|
| Token Telemetry & Quota Gate Sub-ms stream ingestion in Go (TokenGoblin · TokPulse) |
Kernel Tenant Isolation Session JWTs bound in PostgreSQL (Settler v1.8.0) |
Ambient Host Surface Deterministic eBPF ZK-VM sandboxing (zerokernel v1.0.0) |
Formal Verification Lean 4 kernel proofs & Quint specs (GRAPH-Lab · veridag) |
CI-Gated Production TagsSettler, FORGE, zerokernel, ControlPlane, TokPulse, SAT-Frontier |
| System Domain | Architectural Approach & Enforced Invariants | Shipped Repositories & Releases |
|---|---|---|
| AI Platform, Evals & Agent Governance | Keep foundation models unprivileged; enforce MCP capability contracts, offline regression/grounding assertions (Context Caching & Thinking Tokens), and policy-gated delivery. |
ReadyLayer TS · FORGE v1.0.0 · AgentAxiom Py · Vertex/Gemini · agent-infra |
| Multi-Tenant Data, RAG & Kernel Isolation | Enforce tenant and sub-document RBAC boundaries in PostgreSQL (FORCE RLS), verify SHA-256 RAG citation provenance, and isolate untrusted code in ZK-eBPF and WASM sandboxes. |
Settler v1.8.0 · TS · zerokernel v1.0.0 · Rust/eBPF · WorldVM Rust/Wasmtime · RAG Verifier |
| Inference Economics, FinOps & GPU Roofline | Measure token throughput, tail latency, and spend in sub-millisecond streams; compile PagedAttention KV-cache & quantization topologies against hardware rooflines. | TokenGoblin Go · TokPulse v3.0.0 · Go · KV Roofline Compiler Py/Triton |
| Deterministic Distributed & Native Execution | Specify causal state transitions, PostgreSQL exactly-once RPC queues, and native C++/Rust runtimes validated against Quint formal models. | veridag Rust/Quint · ControlPlane v1.0.0 · Rust · Requiem C++/TS · JobForge Postgres/RPC |
| ML Compilers & Numerical Fidelity | Turn silent floating-point discrepancies across large JAX/OpenXLA computation graphs into minimal, portable, deterministic reproducers. | XLAxiom Python · JAX/OpenXLA |
| Formal Methods, SMT & Enterprise Standards | Pair exact rational/combinatorial search, QF_NRA SMT campaigns, and SAT solving with Lean 4 kernel witnesses (Zero Axioms) and signed EdTech (CASE) ledgers. |
GRAPH-Lab Lean 4/Py · UNIFORM Rust/Lean 4 · SAT-Frontier v0.2.9 · Cred-bridge CASE |
Production AI requires strict deterministic boundaries around probabilistic models. The model proposes candidate intent; typed policy services authorize what may happen; isolated kernels execute only approved work; cryptographic ledgers make every outcome inspectable and replayable.
| Production Failure Mode | Why Naïve Agent Wrappers Fail | Hardonian Kernel / Protocol Mitigation | Verified Implementation |
|---|---|---|---|
| Prompt-injected cross-tenant read/write | Application-layer WHERE tenant_id = ? filters are easily omitted by generated queries or bypassed by ORM bugs. |
Session-scoped JWT claims bound directly in the PostgreSQL kernel via FORCE ROW LEVEL SECURITY, verified by automated negative tests. |
Settler (v1.8.0) · RLS Architecture Note |
| Unauthorized tool execution & privilege creep | Passing raw API credentials or broad tool schemas to a model lets an injected prompt trigger destructive side effects. | Unprivileged planner emits typed intent; an external MCP policy gateway validates capability scope and mints single-use approval tokens. | ReadyLayer · MCP Authorization Note |
| Runaway agent loops & unmeasured token spend | Multi-step retries and recursive tool calls compound tail latency and exhaust budgets long before billing alerts fire. | Sub-millisecond token stream ingestion, tenant-level hard quota circuit breakers, and deterministic fallback model routing. | TokenGoblin · TokPulse (v3.0.0) |
| Out-of-order distributed state mutation | Concurrent workers and network retries violate causal ordering when state transitions are implicit. | Quint state-machine specifications, capability-checked causal DAG transitions, and ZK-proved eBPF sandboxes. | veridag · zerokernel (v1.0.0) · ZK eBPF & DAG Note |
| Silent compiler numerical divergence | Floating-point discrepancies inside multi-thousand-node JAX/OpenXLA graphs are intractable to isolate by hand. | Automated HLO graph delta-minimization that bisects and extracts minimal, portable numerical reproducers. | XLAxiom · GPU Roofline Note |
| RAG role leakage & hallucinated citations | Post-filtering ANN vector search leaks restricted chunks across roles, and prompt-level citations can be fabricated. | Pre-filtered FORCE RLS + BM25/pgvector RRF fusion paired with cryptographic [cite:id@sha256] span verification. |
AgentAxiom · RAG Verifier · RAG Provenance Note |
| Unverified algorithmic or mathematical bounds | Heuristic search scripts can misreport extremal counterexamples without machine-checked proof obligations. | Exact rational/SMT search and SAT solving paired with Lean 4 kernel-verified proof witnesses (Zero Axioms). |
GRAPH-Lab · UNIFORM · SAT-Frontier (v0.2.9) |
These four canonical repositories represent how I design, test, and verify software. The table below is generated from a versioned manifest, audited against live GitHub endpoints in CI, and separates stable, beta, and research maturity tiers.
Public project metadata last verified 2026-11-07 · source manifest · verification policy
| Project | Problem | Public evidence |
|---|---|---|
| TokenGoblin · Go Measure · beta |
LLM workloads need cost, usage, and routing data before teams can control inference spend. | Public ingestion benchmarks, cost and routing tests, and a repository-level CI workflow. Architecture · Evidence |
| ReadyLayer · TypeScript Govern · beta |
AI-assisted delivery needs policy, review, and evidence before generated changes reach production. | Public policy contracts, evidence export documentation, test suites, and CI quality gates. Architecture · Evidence |
| veridag · Rust Prove · research |
Distributed execution needs explicit ordering, capability security, and cross-language conformance. | A public protocol specification, Quint formal models, test vectors, and dedicated conformance workflows. Architecture · Evidence |
| Settler · TypeScript Reconcile · stable · v1.8.0 |
Payment, banking, and operational records diverge unless matching and evidence rules are explicit. | Multi-tenant reconciliation engine v1.8.0, 20+ CI gates, public benchmark suite, and cryptographically verified release artifacts. Architecture · Evidence |
Architecture monographs and reference specifications grounded directly in the production systems above:
| Monograph / Specification | Core Architectural Thesis | Companion Codebase |
|---|---|---|
| 01 // Building Safe Multi-Tenant AI Agents with Postgres RLS | Why application-layer WHERE filtering fails in autonomous workflows, and how to bind session-scoped JWT tenant claims directly to PostgreSQL FORCE ROW LEVEL SECURITY. |
Settler (v1.8.0) |
| 02 // Enterprise Tool Calling: Authorization Outside the Model | Decoupling probabilistic planning from privileged execution via policy-gated MCP reverse proxies and single-use cryptographic approval tokens. | ReadyLayer · FORGE (v1.0.0) |
| 03 // From Prototype to Production: Enterprise AI Agents | End-to-end production hardening across model routing abstractions, deterministic state boundaries, CI evaluation gates, OpenTelemetry tracing, and FinOps circuit breakers. | TokenGoblin · TokPulse (v3.0.0) |
| 04 // Zero-Knowledge eBPF Sandboxing & Causal DAG Execution | Formalizing causal state-machine preconditions in Quint, isolating untrusted execution inside deterministic eBPF ZK-VMs, and emitting replayable SHA-256 evidence envelopes. | zerokernel (v1.0.0) · veridag · ControlPlane (v1.0.0) |
| 05 // Permission-Aware Hybrid RAG & Cryptographic Citation Provenance | Pre-filtered PostgreSQL FORCE RLS + BM25/pgvector Reciprocal Rank Fusion (RRF) and deterministic [cite:chunk@sha256] claim grounding verification. |
RAG Verifier · AgentAxiom · Settler (v1.8.0) |
| 06 // GPU KV-Cache Roofline Modeling & Deterministic Serving Topology | Modeling GQA PagedAttention KV-cache footprints, memory-bandwidth vs. compute rooflines, and Triton fused INT4 dequant-GEMV kernels across heterogeneous GPUs. |
KV Roofline Compiler · XLAxiom · TokenGoblin |
| 07 // Hardonia Sovereign AI Architecture Playbook | System topology, bounded contexts (Observe → Control → Operate → Prove → Reconcile), verified release tags, and cryptographic evidence boundaries. |
Hardonian · Executive Brief |
I operate a dedicated hardware lab for local model routing, quantized inference benchmarking, multi-GPU failure-mode injection, and telemetry soak testing:
| Compute Tier | Hardware / Runtime Specification | Public Tooling & Workloads |
|---|---|---|
| Server & Accelerator Rack | AMD EPYC 7452 (32C/64T server host) · NVIDIA Tesla V100 / P40 / RTX 3060 multi-GPU array | Quantized LLM serving, OpenXLA compiler bisecting, KV Roofline Compiler |
| Edge & NPU Workstation | AMD Ryzen AI 9 HX370 (XDNA 2 NPU + Radeon 890M) · aarch64 / Linux container parity |
Local agent control loops, low-latency MCP gateway testing, RAG Citation Verifier |
| Routing & Observability Plane | Ollama-compatible model routers · Docker/OCI · Prometheus & OpenTelemetry collectors | model-tools · autopilot · api-tools |
Reproduce & verify this profile locally
Every project claim, live CI workflow badge, local SVG asset, and external link in this README is continuously verified by automated test suites:
git clone https://github.com/Hardonian/Hardonian.git
cd Hardonian
python -m unittest discover -s tests -v
python scripts/profile-metadata.py --check --verify-remote
python scripts/profile-link-audit.pyI am a Solutions Architect at McGraw Hill and build the Hardonia ecosystem independently from Toronto. My career combines enterprise-scale technical strategy with hands-on systems engineering: designing distributed control planes, formal state models, multi-tenant TypeScript/Rust/Go services, and database-kernel security boundaries.
I also contribute specialized part-time expertise to confidential frontier-AI evaluation and systems initiatives (client, model, dataset, and internal research details remain strictly private).
- Enterprise Architecture & Integration: Deep field experience integrating identity providers, event buses, and institutional platforms (
OAuth 2.0,Keycloak OIDC,SCIM,LTI 1.3 Advantage,OneRoster 1.2,MCP, andOpenTelemetry). - Security & Governance: Proven track record leading institutional security, data residency,
SOC 2, andFERPAcompliance architecture reviews. - Commercial Impact: Recipient of the President's Award for Sales Excellence, reflecting the ability to translate rigorous systems engineering into customer trust and enterprise revenue.
- Target Roles: Senior / Staff Applied AI Engineer · AI Platform Engineer · Distributed Systems Engineer · Principal / Solutions Architect.
| Audience | Scope & Focus | Direct Channel |
|---|---|---|
| Hiring Leaders & Executive Recruiters | Senior / Staff / Principal Applied AI Engineer · AI Platform Engineer · Distributed Systems Engineer · Principal Solutions Architect | LinkedIn · Email · Executive Brief |
| Founders, Investors & Technical Advisory | AI control-plane architecture reviews, multi-tenant database isolation audits, LLM FinOps governance, and technical due diligence | LinkedIn · Email · Architecture Playbook |



