I build verifiable AI and cryptographic systems and prove they work outside the lab. Most recently: zero-knowledge and post-quantum verifiers executed in Bitcoin Script on the BSV mainnet.
$ whoami Gregory J. Ward Research Engineer · CTO @ SmartLedger $ cat focus.json { "research": [ "axiomatic models", "LLM training", "agent evaluation", "reinforcement environments" ], "security": [ "tamper detection", "cryptographic integrity", "sandboxed AI agents" ], "systems": [ "distributed computing", "agentic AI", "verifiable infrastructure" ], "cryptography": [ "post-quantum signatures", "pairing + SNARK verifiers in BSV Script" ] } $ _
My work sits where cybersecurity, artificial intelligence, cryptography, distributed systems, and experimental computation meet. I build systems that do more than produce outputs — they preserve evidence about how those outputs were produced.
Four examples of the work I do: take a problem most people treat as impractical, build it end to end, and leave evidence anyone can check.
Selected milestones, dated by when each project began, often well before the wider industry got there.
Each LLM answer triggers a sub-cent BSV micropayment, weeks before the ChatGPT API launched on March 1.
GPT-3.5 writing sCrypt contracts with retrieval over the full sCrypt source, early in the LangChain wave.
Hash chains signed with SPHINCS+ and ML-DSA, within months of NIST finalizing FIPS 204 and 205.
LLM-written code in a small stack language, checked by a static verifier before it is allowed to run.
A Litecoin fork with post-quantum signatures and multisig at consensus, while Bitcoin was still debating how to migrate.
Delta-sigma trit-stream weights that turn ternary precision into a dial set at inference time, extending BitNet b1.58.
A QUBO / Ising solver that emits optimality certificates checkable with plain arithmetic, applied to protein side-chain packing.
BLS12-381 pairing and Groth16 verification in layer-1 Bitcoin Script, confirmed in blocks 966795 and 966923.
Current work explores model reasoning, autonomous agents, reinforcement-style environments, continual adaptation, and computational discovery.
Since December 2025, I have been investigating models that reason from explicit axioms, invariants, constraints, and falsifiable propositions rather than relying exclusively on unconstrained natural-language inference.
Experimental work with outcome-based reward, persistent memory, self-evaluation, sparse expert routing, localized learning, and reinforcement-style feedback loops.
Testing whether useful parameter-update structure can be retained, compressed, and reused to accelerate later model adaptation across tasks and domains.
Building systems that generate hypotheses, preserve invariants, run controlled experiments, search for counterexamples, and revise their theories from evidence.
My security work has evolved from file integrity and cryptographic event chains into AI-agent accountability, sandboxed execution, and independently verifiable process history.
Formal foundation in defensive security, risk, cryptography, network security, identity, and secure architecture.
Designed a system where file changes become signed fingerprints, are hash-chained, rolled into Merkle roots, and anchored for independent verification.
Led the IBM engineering team during the first implementation of CertiHash and presented the architecture publicly in London.
Current work includes sandboxed terminal agents, Kali Linux evaluation, constrained action spaces, permission boundaries, tamper monitoring, and agent accountability.
In August and September 2026 I ran falsification tests against my own earlier model-training ideas and rejected several of them. I keep the negative results on record alongside the positive ones.
Signed histories, on-chain anchors, equivalence proofs and certificates a third party can check without trusting me.
From leading the IBM team on the first CertiHash implementation to verifiers executed on mainnet, the work goes past the prototype.
“Don’t just ask whether an intelligent system produced the right result. Ask what it did, what evidence it used, what authority it had, and whether the process can be verified.”
I teach programming, artificial intelligence, prompt engineering, agent systems, blockchain, and computational thinking to beginners, developers, organizations, and technical teams.
Programming fundamentals through advanced systems thinking for kids, adults, and working developers.
Practical AI usage, prompt design, agent workflows, model integration, automation, and responsible deployment.
Available for lectures, workshops, panels, research discussions, and technical strategy sessions.
I’m open to research-engineering and technical-lead roles in AI, security and applied cryptography, and to consulting, teaching and speaking.
Research Engineer
CTO / CDO, SmartLedger
Founder, Codenlighten
Opens your email app with the message pre-filled.