Solution architect for AI and blockchain — AI systems that stay affordable in production, and ledgers that stay correct under attack. Toronto, Canada · Axieo Technologies Inc. · [email protected]

I design and build systems where mistakes are expensive: AI applications whose cost grows with every call, and ledgers, payment flows and shared records between organisations that have to stay correct under adversarial conditions. I write them so that other engineers can read them. I work through Axieo Technologies Inc., based in Toronto.

• Consulting — an architect on call: design reviews, model, platform and vendor selection, "do we need an LLM for this?" and "do we need a blockchain?" answered honestly, due diligence, and support during delivery. By the day or on a monthly retainer.
• Solution architecture — from a business problem to a system design a team can build: data model, trust boundaries, failure modes, deployment topology on Google Cloud or AWS, operations, cost.
• AI cost review — where your LLM spend actually goes: prompts, model choice, caching, retrieval and output size, ranked by savings, with a plan to cut the bill without losing quality.
• Private and consortium ledgers — a shared, tamper-evident record that several organisations write to and none can quietly rewrite, without a public token or a heavyweight framework. Fixed price, with a runbook and operator training.
• Architecture review — a written report on an existing system: what can go wrong, ranked; what to fix first; what is fine and why.
• Training — a self-paced course where engineers implement a real blockchain's rules with its own tests as the judge, and live workshops for teams.

tokensaver is an open-source command-line tool that shows what a prompt costs on 20 OpenAI, Anthropic and Google models, and exactly which changes make it cheaper. Each suggestion comes with its reason and the tokens and dollars it saves.

• Runs fully offline: the prompt never leaves the machine, and the tokenizer files ship inside the package, checked against their published hashes
• Exact counts for OpenAI models; Claude, Gemini and GPT-6 shown as clearly labelled estimates, never as exact
• Six rule-based optimizers (JSON minification, duplicates, repeated instructions, filler, whitespace, long few-shot examples). Safe changes are applied by default; anything that could change meaning waits for review
• Every saving is measured before and after on the chosen model: the bundled example prompt drops by 20% on GPT-4o
• Python, 149 tests, CI on Windows, macOS and Linux

The rules are heuristics, and the README says exactly what is exact and what is estimated.

PlainChain is a complete Layer 1 blockchain node written from scratch in TypeScript — small enough to read in an afternoon. No blockchain SDKs: every hash, signature and Merkle tree comes from Node's built-in crypto, and there are three runtime dependencies.

• Proof of work with fork-aware retargeting, or a Clique-style proof of authority for consortium chains
• UTXO ledger with crash-atomic reorganisations, header-first sync, checkpoints
• Replace-by-fee mempool; wallet with BIP-39 phrases, watch-only files and an SPV light client
• JSON-RPC over TLS, Prometheus metrics, Docker and Kubernetes packaging — deployed and tested on Google Cloud (GKE)
• 622 tests, a large share of them explicit attacks; a public threat model covering about 60 threats, each tied to code and a test

Built test-first over four stages with the engineering conventions documented in the repository. Open source. Not yet independently audited, and the repository says so.

Write to [email protected] with a paragraph about the problem. You will get a first opinion within two business days — including "you may not need this".