Blackwell Systems is the practice of Dayna Blackwell. I build production systems and publish the proof, then bring that same evidence-first standard to the technical decisions other teams need to get right.

The spine under all of it: independent, evidence-first technical judgment. The outside read a team brings in for the calls that are hard to make from the inside. Is the claim real, should you build or buy, what should you own, and what is it costing you. Backed by published work and verifiable on your own systems, not accepted on authority.

Fractional CTO

Embedded technical leadership for teams that need the judgment before they hire a full-time CTO: strategy and roadmap, build-versus-buy and vendor decisions, architecture, engineering process, cost, and the technical narrative for a raise. Delivered as a decision memo with the criteria, the basis, and what would change the recommendation, the same shape as the diligence work below.

Technical Due Diligence

Independent evaluation of a claim or a vendor: is the advantage real, and should you buy, build, or own it.

Most technical due diligence is demonstrative: the founder shows you it works. Mine is adversarial: I define what would prove it doesn’t, then try. I establish independent ground truth where possible, compare against competent baselines, and measure scaling behavior rather than curating benchmark wins. Then I mark the boundary where an advantage holds, degrades, or becomes unknown.

Claim types: speedup and scaling claims, token and inference-cost advantages, determinism and reproducibility, novel computational primitives, model and preprocessing attribution, and claims of superiority over existing methods. Engagements run under NDA, proprietary systems stay inside your environment, and findings are independently reproducible.

AI Cost Optimization

Cut the LLM bill without losing quality. A full-stack audit across techniques that compound to 70-90% savings, every number measured on your data.

Not sure where AI fits yet

The AI Opportunity Assessment is a fixed-scope, vendor-neutral entry engagement for owners and operators: where AI fits your business and your constraints, what to decline, and what it would cost. Nothing to sell, no reseller margin, no affiliate links, which is the one thing most “neutral” advisors cannot say.

Selected work

The judgment above is not theoretical. It runs on the same standard as the open-source infrastructure the AI ecosystem is building on. Every figure here is measured and reproducible.

GCF is an AI-native wire format reverse-engineered from tokenizer data. 100% comprehension on every frontier model on standard workloads, 91.2% on structurally complex code graphs where JSON drops to 54.1%. 50-92% fewer tokens than JSON. 43 billion+ lossless round-trips across 5 formats with zero failures. Seven language implementations, seven registries, a tree-sitter grammar. Used by 28 projects outside Blackwell Systems, with 580K+ downloads, including Chrome DevTools MCP (the #1 MCP server on GitHub), OmniRoute, Speakeasy, the Linux Foundation’s Open Data Products SDK, netclaw, ctx, and NeuroNest. Five published whitepapers.

knowing is a content-addressed code intelligence engine that beats every competitor in the category with statistical proof. P@10 of 0.278 across 308 tasks, 16 repos, and 8 languages: 3.2x codegraph, 5.05x GitNexus, 5.35x Gortex, 12.1x Aider, 18.5x grep. 28 MCP tools, 23 extractors across 26 languages, supply-chain detection without executing code at a 1.0% false-positive rate. Single Go binary. Published whitepaper with DOI.

agent-lsp is a stateful MCP runtime over real language servers: 66 tools, 24 Agent Skills, a speculative execution engine, 30 CI-verified languages. Listed on the official MCP Registry, Glama (A-tier), and awesome-mcp-servers.

mcp-assert is the deterministic testing standard for MCP servers. 55,000+ downloads across 6 channels, 102 servers scanned, 34 upstream bugs found, including 9 crashes in Ant Group’s mcp-server-chart, fixed upstream. Runs in CI across 39 of WYRE’s MCP server repos.

polywave is a formally specified parallel-agent coordination protocol: 6 invariants, 48 execution rules, 7 roles, 4-5x measured speedup. knowing (94K lines) was built on it.

GCP Emulator Platform: five composable emulators (Secret Manager, KMS, IAM, Eventarc, auth). The Secret Manager emulator is the most widely adopted community solution, ranked #1 on Google, Bing, and DuckDuckGo, with 275K+ downloads, used in the test suites of Apereo CAS, Camunda and Flipt among 10 projects using the emulators.

Across the ecosystem: 20+ open source projects used by 120+ repositories across 60+ organizations, 1M+ downloads, 50+ upstream PRs merged into Google, Anthropic, GitHub, Grafana, and HashiCorp, and 11 self-published research papers that drove inbound. The judgment I sell is the judgment I use.

Browse the full open source portfolio →

Work with me