Platform · cloud · agentic systems

The practice you retain when the platform is the problem.

SoulTech Solutions is a boutique software engineering practice. Engineering leaders retain us to review architecture before a year gets committed to it, to take monoliths and proofs of concept apart into service fabric, to build the developer platform that carries complexity instead of the team, and to get agentic systems through production.

The practice
Availability
Currently taking new engagements for Q4 and Q1.
Delivery
Principal-only, hands on the code. No bench, no handoff.
Engagements
Assessment · 2–6 weeksProject · 3–12 monthsRetainer · days per month
Based
Ravensdale, Washington · remote-first, Pacific time
Who retains us

VPs of Engineering and CTOs at scaling startups · founders whose proof of concept has to become a product · heads of platform and developer experience at mid-size orgs · executives and boards who need a technical translator · investors who need a real read before the check.

01 — Services

What SoulTech gets retained for

Nine shapes of work. Each one is scoped to a decision you need made and an outcome you can verify — and each one ends with your team able to operate what was built.

See how each engagement works →
02 — Engagement shapes

Three ways to retain the practice

Assessment
2–6 weeks · fixed scope

A decision, written down.

Architecture review, AI and platform readiness, or technical diligence. You leave with findings, a target architecture and a sequenced plan you can execute with us or without us.

Project
3–12 months · milestones

Built, operated, documented, handed over.

Decomposition, a developer platform, an execution plane, a hardening programme. Scoped to milestones with a demonstrable outcome at each one, and an operational handover at the end.

Retainer
monthly · days per month

Standing judgment on call.

Fractional principal, interim architect, or continuing advisory after a project. A fixed allocation of days, reserved, with an escalation path when something breaks.

Fees are quoted against the scope, not against a rate card — and an assessment is a legitimate place to stop. NDAs signed before the first technical conversation as a matter of course.

03 — How the work goes

Four commitments, on every engagement

  • 01

    Measure before optimizing

    An estimate of “30 to 90 seconds, plausible” turned out to be 360. The measured figure changed the architectural conclusion, not just the tuning.

  • 02

    One principal, no bench

    The person in the room designing it is the person writing it. No account layer, no staffing pyramid, no quiet handoff to a junior team once the contract is signed.

  • 03

    Cognitive load stays with the platform

    Your engineers should get a capability by including a target, not by learning how it works. Complexity is the platform’s job to absorb.

  • 04

    Own the outcome, then hand it over

    Engagements end with your team operating what was built, mentored on the patterns underneath, with the decisions and runbooks written down.

04 — Agentic engineering

We build agentic systems. We also build with agents.

Claude, Codex and Gemini are part of how this practice works — in planning, architecture review, implementation and dependency triage. It is why one principal can deliver at the pace of a small team, and why the platform work above is affordable at boutique scale.

The difference is accountability: a principal engineer still owns every line that ships. Agent output is reviewed, tested, load-tested and understood before it reaches your repository. You get the leverage without inheriting code nobody can explain.

Where agents sit in the SDLC
01
Planning
Scope, phasing and trade-offs drafted fast, decided by a human, recorded as ADRs.
02
Architecture
Designs stress-tested against their alternatives before code exists; C4 and contracts generated from the decision.
03
Implementation
Scaffolding, tests and infrastructure generated against pinned golden templates, then reviewed line by line.
04
Review & triage
Agent-assisted code review, dependency and CVE triage, and incident forensics against real telemetry.
05 — Selected work

The evidence, in detail

All work →
Case 01 · Seed-stage AI data platform

From contractor proof of concept to a platform an engineering team could build a company on

Twelve concurrent domains over ten months: a developer platform from zero, an eight-phase decomposition, an event-driven execution plane, a measured scaling model, and the security and release engineering underneath it. In production since June 2026.

360s → 5s
Cold-start scale-out
2% → 0.3%
Failure rate
64
Repos on the platform
Open the ledger →
Case 02 · Product of the practice

The Fudger

A three-tier product designed and shipped solo over the past year — marketing surface, single-page application, and API service — now approaching beta. It doubles as the practice’s proving ground: the same golden-path tooling, service design and release automation we build for clients, run end to end on our own product.

Full write-up in progress
06 — Start

Tell us what’s hard. We’ll tell you whether we’re the right practice for it.

The best first conversation is a specific one — a system that will not scale, a launch nobody can vouch for, an AI feature stuck between demo and production, a codebase somebody is about to buy.

Start a project
Response
Within two business days.