AI implementation & systems advisory

AI that earns its place in the business.

I'm Keith Soyka — 22+ years running real-world operations, now building and advising on AI systems that survive contact with reality. Through GestaltView and client work, I help small and mid-sized businesses find the right AI opportunity, build it responsibly, and put it into operation — plus design services that make complex ideas legible.

Operational judgment. Production AI. One accountable builder. · New York / remote · Available for part-time, contract, fractional, and advisory work.

22+years frontline operations leadership — P&L, labor, inventory, multi-site
16months designing & shipping a multi-model AI runtime, solo
217Postgres tables · 299 RLS policies · 75 DB functions (Aug 2026 snapshot)
03engagement tiers — every one ending in documentation & a real handoff
What I do

Three lanes, one standard: inspectable work.

Whether it's an AI system, an operation, or a brand identity, the work follows the same discipline — evidence first, interpretation labeled, handoff documented. Pick the lane that matches your problem.

Lane A — Advisory & Build

AI Implementation & Systems Advisory

Opportunity audits, bespoke agent/system builds, and fractional advisory for teams that need AI sized to the actual constraint — not a stack looking for a reason to exist.

  • AI Opportunity Audit — from $2,500
  • Bespoke AI Agent / System Build — from $5,000
  • Fractional AI Advisory — from $2,500/mo
How engagements work →
Lane B — Design Services

Design that makes ideas legible

Logo and identity systems, pitch decks investors can follow, web design, and software builds — from someone who has iterated a brand through dozens of versions and lived to document it.

  • Logo design — $175 personal / $350 business
  • Presentation design — $75/hr
  • Web design & software builds — scoped
Design services →
Lane C — Operations leadership

Fractional ops for businesses that run on people

22+ years of frontline operations leadership — P&L, labor, inventory, multi-site transitions, training. The same discipline as the systems work: find the constraint, fix the process, document the handoff.

  • P&L and financial accountability
  • Labor planning and team leadership
  • Multi-site operations & transitions
Operations engagements →
Selected work

Built to be inspected, not mythologized.

A production multi-model AI runtime and the systems around it — described with the evidence labels each claim has actually earned.

LiveRuntime

GestaltView v3

A user-directed continuity and recognition system: preserves source context, distinguishes original material from interpretation, and connects confirmed context to action. Multi-model orchestration, hybrid retrieval, governed data handling.

Open the live runtime →
LiveAgent

Billy — collaborator agent

The context engine at GestaltView's center, running across web, Discord, and Reddit. Built around ten constitutional invariants — five user-facing, five for digital intelligences.

How it's built →
In buildEvaluation

Portfolio Throughline Test

An eval harness that checks whether an AI system preserves relationships across tasks — five local tasks, a scored rubric, contradiction and missing-evidence probes. The methodology I bring to client evaluations.

The methodology →
Evidence standard: every claim on this site is labeled by what backs it — verified (artifact, test, or observation), staged (in build, documented), or planned (intent, not yet built). Nothing here asks for trust it hasn't earned.
Field note
"The audience sees light. The operator sees the chain of small decisions that makes the light possible."
— Years in projection booths and frontline operations taught me to pay attention to consequences: the invisible choice, the failure mode, the handoff path. That's the same lens I bring to AI.