Design, Measure, and
Optimize LLM Systems
Rent the expertise. Own the architecture.
For engineering teams putting LLMs at the core of their product — whether you're designing your first system or fixing one that's underperforming.
01 / LLM expertise — delivered to small teams
The engineers who've shipped production LLM systems at scale are locked up at frontier labs and Big Tech, at seven-figure comp. You can't hire them — and you don't need to. Smartfast packages that expertise into something a small eng org can actually buy: a fixed scope, a measured result, and a design your own team builds.
02 / How It Works
- 02.a
Consult — 30 min, $150.
The first-pass filter, not a sales call. You bring your product, your system (if one exists), and what's not working; I bring pattern-matching from years of building these systems. You'll leave with my honest read. I keep a small client roster and a paid front door; if $150 gives you pause, we're not a fit yet.
- 02.b
Working session — 1 hr, $300.
We go deep on your project: architecture, data, constraints, options, tradeoffs. Real work happens in the room, not discovery theater — and we end with a decision: is this worth a full assessment?
- 02.c
Assessment — $3,500.
This is where I really dig in. We sign the paperwork, I get access — code, data, knowledge bases — and I go deep offline. You get a written assessment that defines the problem precisely, judges feasibility, and flags the risks — including "this isn't a good fit for LLMs" if that's the truth. It's your first real checkpoint, and it includes the working session where we go through it together.
- 02.d
Proposal — no cost.
If we both want to proceed, I come back with a fixed-scope proposal: scope, deliverables, timeline, cost, and initial targets, in writing. Targets at this stage are calibrated guesses — the checkpoints are where evidence turns them into real numbers. We iterate until it's right — nothing starts until we agree.
- 02.e
The work — from $25,000.
The core of the engagement, and it starts with data: I build the training and eval datasets first, so every decision after has numbers behind it. Then the loop — I iterate on a working prototype, run the evals, and we meet weekly to go over progress, challenges, and direction. When the evidence holds, I distill the prototype into a concrete development plan, so your production build sheds all the R&D tech debt. Depending on scope, the prototype is a full end-to-end implementation or validates the riskiest components piece by piece.
- 02.f
Handoff — everything transfers.
The prototype codebase, the design doc, the eval suite and the datasets behind it, benchmark results, observability dashboards, prompts, scripts, tooling — all of it. If I made it during the engagement, you own it. Your team — or your contractors — build production.
- 02.g
I stay reachable.
The design doc answers most questions; for the rest, I'm a message away. Questions during the build land on my desk, not in a black hole.
03 / Checkpoints — honest assessment, built in
Targets are guesses. Measurements aren't.
Some goals can't be judged from the outside — figuring out whether a thing is feasible is itself real work. That's why every proposal has checkpoints scoped in: fixed points where we stop, put the numbers on the table, and make an honest call — push on, change course, or stop.
A checkpoint isn't a status meeting. You get the eval results, the benchmarks, and my written read of where we stand against the KPIs we agreed to. If the evidence says the goal is reachable, we keep going. If it says otherwise, you hear that instead — and everything produced so far is yours either way. No sunk-cost momentum, no polite optimism.
A big part of what you're paying for is a straight answer: "I don't care what Sam Altman said — that isn't possible." You'll hear it the moment I know it, not after the budget is spent.
One price, paid at checkpoints. If the evidence says stop, you stop paying — and you keep everything.
04 / Why This Model
You don't need the $1M hire.
A $1M-a-year AI engineer doesn't make sense for a small eng org. Renting the expertise for a fixed scope does — and not just because of the price tag. This model has very specific advantages:
- ✓No onramp. I design your system — I don't have to learn your entire codebase to do it. Week one is productive work, not months of archaeology billed at expert rates.
- ✓You pay top dollar for expertise only. Design and verification — not for someone to manage an LLM while it writes code, productionizes, and deploys. Your team does that work at your rates, in your stack, to your standards. Implementation gets dramatically cheaper.
- ✓Structural wins, not marginal ones. We focus on the big architectural moves — a future-proof, scalable, maintainable, observable system — not squeezing out 3% by piling on more technical debt.
- ✓Your team levels up. They build it, so they own it — they can run it, extend it, and debug it without me. When I leave, the expertise stays in your codebase, not in my head.
05 / Next Step
Bring me your system — or your idea. In 30 minutes we'll figure out where it stands, what it'll take, and whether we're a fit. You'll leave with my honest read either way.
06 / About
Hi! I'm Todd Sifleet. I've spent 15+ years building software — principal engineer at Pathwork Life, tech lead at Uber, founder and CTO of Litlingo, engineer at Athenahealth — and I've worked with LLMs since their inception, including founding two companies in the space.
I don't just talk about this — I build with it every day.
Read my writing on Substack →