Services: fractional CTO, technical due diligence, and AI cost optimization
How I work
One method, several ways in
One method underneath everything: evidence-first, verifiable, reproducible. Every recommendation is backed by published work and checked on your own data, not asserted. Start with an assessment, or go straight to the service you need.
Fractional CTO
Embedded technical leadership: strategy and roadmap, build-versus-buy and vendor decisions, architecture, engineering process, cost, and the technical narrative for a raise. The judgment a company needs before it hires a full-time CTO.
Technical due diligence
Adversarial, not demonstrative: I define what would prove a claim false, then try. Independent evaluation of a technical claim or vendor, is the advantage real, and should you buy, build, or own it, delivered as a decision memo with economics and risk.
AI cost optimization
Cut your LLM bill without losing quality. A full-stack audit across ten techniques that compound to 70-90% savings, every number measured on your data.
Run a business and just need a straight answer on AI?
If you are not technical and do not need a CTO, start with the AI Opportunity Assessment: a fixed-scope, vendor-neutral read on where AI actually fits your business and where it does not. Plain language, nothing to sell you.
Fractional CTO
The judgment before the full-time hire
Brought in as a service, a few hours a week or for a defined project. CTO-level decisions without the full-time cost, with clear scope: I own the technical calls and guide your team, and I am direct about what is decision work versus hands-on build.
Strategy and roadmap
Align the technical direction to the business, set the roadmap, and make the big architecture calls.
Build, buy, or own
Evaluate technologies and vendors, run build-versus-buy, and decide what to own versus rent.
Architecture and scale
System design, reliability, security posture, and paying down the tech debt that slows delivery.
AI systems and cost
AI strategy, model and vendor selection, and keeping the AI bill sane as you scale.
Team and process
Engineering process, hiring guidance, and leveling up the people you already have.
Fundraising support
The technical narrative for a raise, and answering investor technical due diligence from the technical seat.
Technical due diligence
Is the claim real. Should you buy, build, or own it.
Most diligence is demonstrative: the founder shows you it works. Mine is adversarial: I define what would prove the claim false, establish independent ground truth, and try. Independent, no stake in the vendor, every finding traces to evidence and reproduces.
Claim verification (the engine)
For computational-advantage and compute-primitive claims: independent ground truth, classical-solver baselines, and a battery covering attribution, determinism, mechanism, scaling, fair comparison, and correctness. Does the advantage actually hold.
Vendor evaluation (the method)
For a buy-or-build decision on a product or vendor: the real peer class, the same-task comparison the vendor avoids, mechanism versus model, and the own-versus-rent economics.
What you get
A decision memo with a clear recommendation, a rent-versus-build cost model, a risk register, and a methodology note stating scope and confidence. Written to be acted on.
Independence
Public materials, no stake, and a standing commitment to revise on new evidence. The assessment is worth paying for because it is not sold to a conclusion.
Worked example
A full public teardown of an AI decision-layer vendor: the peer class its benchmarks avoid, measured head-to-head results, and the build-versus-buy call. When new experiments reversed part of my own conclusion, I published the correction.
AI cost optimization
10 techniques that compound multiplicatively
Most tools optimize one layer. I diagnose which layers matter for your pipeline and implement the full stack.
Prompt Caching
Restructure prompts for provider-native caching. 90% savings on cached tokens (Anthropic), 50% (OpenAI). Most teams leave this unconfigured.
Wire Format Compression
GCF encodes structured data with 61-71% fewer tokens than JSON. Drop-in proxy for zero-code adoption, or native integration for session dedup.
Model Routing
Route simple tasks to cheap models, complex tasks to expensive ones. 50-80% cost reduction depending on your query mix.
RAG Optimization
Rerank before injection, tune chunk sizes, encode retrieval results. Most teams retrieve 10-20 chunks when 3-5 would suffice.
Batch API
50% discount on any workload that can tolerate 24-hour latency. Classification, extraction, content generation. Free money.
Agent Loop Optimization
Batch tool calls, progressive disclosure, reasoning effort control. Reduce 15 tool calls to 5 with better tool design.
Session Deduplication
Track what data the model has already seen. GCF session dedup makes call five 92% smaller than call one.
Prompt Engineering
Stable prefix design for cache hits, system prompt compression, instruction dedup, few-shot pruning. 30-50% savings from a one-time audit.
Semantic Caching
Cache responses for semantically similar queries. 100% savings on cache hits. GPTCache, Upstash, or custom.
Output Compression
Control reasoning effort via provider APIs (30-70% fewer output tokens). Structured output constraints. Response format optimization.
Multiplicative, not additive
How the savings compound
Base cost: $1.00 per session
Your current cost with no optimization. JSON payloads, no caching, one model, full context on every call.
After prompt caching: $0.40
60% of your tokens are the cached prefix (system prompt, tool schemas). Anthropic charges 10% on cached tokens.
After wire format compression: $0.16
GCF reduces the remaining structured data by 61%. Tool responses, RAG results, batch payloads.
After model routing: $0.05
70% of calls routed to a model that costs 80% less. Same quality on simple tasks.
After batch API: $0.03
Async workloads moved to batch processing at 50% discount. Your optimized cost per session.
Engagement model
Start with a demo. Go as deep as the savings justify.
Free Demo (30 min)
We run gcf-proxy against one of your MCP servers, live, on your data. You see per-call token savings in real time. Walk away with a number. Zero commitment.
Token Cost Audit
Full pipeline analysis across all 10 techniques. We instrument your pipeline, measure token flow, and deliver a report with per-source savings, prioritized roadmap, and 12-month cost projection.
Implementation
We deploy the optimizations from the audit. Prompt restructuring, proxy deployment, model routing, RAG reranking, agent loop redesign. Measured before/after on every change.
Native GCF Integration
For teams that want maximum savings. GCF library integrated into your serialization layer, session dedup and delta encoding across agent loops, conformance tests against your data. Unlocks 92%+ savings on multi-turn sessions.
Why Blackwell Systems
Evidence you can check
Built the format
GCF is our format. Seven implementations, formal spec, 43 billion+ lossless round-trips across 5 formats and 7 languages.
Proved it works
2,500+ LLM evaluations across 11 models and 4 providers. 100% comprehension accuracy on every frontier model. Published methodology, published logs.
Independent by default
The recommendation is not sold to a conclusion. I advise prompt caching, model routing, and batch APIs even though I do not sell them, and diligence is run from public materials with no stake in the vendor.
Bring in an outside technical read.
A short call, or send the details. Independent, evidence-first, and direct about what is worth building.