Engineering deep-dive · every figure sourced

One operator. One agent fleet.
~424,000 lines of real code.

This is the engineering story behind CloviTek: how one founder directs a fleet of specialized AI build-agents to research, design, build, verify, and ship software — and the knowledge layer that lets every build make the next one faster. The build-cost and line-count figures below are proof points from our own ledger, not a pitch-deck estimate.

Measured Estimate Indicative
The numbers, measured

What the fleet has actually produced

Pulled from the build ledger and a live sweep of our own repositories — not round numbers. Token counts and build-minutes are measured per-agent telemetry; dollar figures are tagged estimates or accounting-grade totals as noted.

0
AI agents built & logged
ledger
src: agent build ledger
0
Lines of our own code
repo sweep
src: cloc / find sweep, own repos only
0
Distinct products shipped
live probe
src: HTTP probe of public product URLs
0
RAG / knowledge-base docs indexed
live count
src: embedded index entry count
What it took to build

$57K to build the whole factory

Not a fundraising estimate — the real, reconciled all-in cash spend, from the filed LLC P&L and the founder's own card statements. Presented here as a build-cost proof point, not a per-unit cost breakdown.

2025 — booked (LLC P&L)

$13,843

Filed opex across tooling and subscriptions for the year. accounting

2026 — card, year-to-date

$15,694

Measured from card statements; still growing as the year continues. card statements

Cash to date

~$57K

2025 booked + 2026 card year-to-date — a bootstrapped build, not a funded burn. measured

Cash + founder time

≈ $200K

Adds an estimate of founder sweat-equity on top of the cash figure above. time estimate

The codebase

424K lines, counted transparently

Own code only, across named repository roots — vendored libraries and dependencies are excluded and never claimed as ours.

424K lines across our own repositories
indicative sweep
TypeScript Python JavaScript HTML/CSS Shell

Counted live over named repo roots, excluding node_modules / dist / vendored dependencies.

A self-improving knowledge layer

Agents read from it. Agents write back to it.

Not just code — measured, embedded knowledge bases the fleet consults before it builds and updates after it ships.

41
UI/UX knowledge-base rules
41
RAG docs embedded & indexed
65
Agent builds in the ledger
local
Embedding model — zero marginal cost
How the factory builds

Every build moves through the same gated pipeline

An idea is scoped, built by specialized agents, verified against a quality gate, and shipped — the same loop that produced this page.

How the fleet is wired

One operator, specialized agents, a shared knowledge layer

A simple architecture: an orchestrating operator dispatches specialized build-agents; agents read the shared knowledge layer and emit products and data — which feed back into that same layer.

Operator
Founder + AI orchestratordispatch · review · ledger
Build agents
researchdesignbuildverifydeploylearning loop65 logged
Knowledge
UI/UX knowledge baseproduct corpusoptimization KBbuild ledger
Output
live productsstory pagesdata artifacts→ feeds knowledge back
Platform architecture

Three layers powering 63+ products

Products

63 distinct products & platforms built on the same engine

Legal toolsAccessibilitySEO & contentCreative…and more
Engine
Design token engineBenchmark harnessAgent orchestration
Infrastructure
Cloud VPSObject storageCDNMySQLRedisProcess orchestrator

The stack, as detected

Real, currently-running technologies powering the fleet — no vaporware.

Node.jsPythonTypeScriptMySQLRedisffmpeg Enterprise voice engineEdge worker platformLLM cascade layer Next.jsExpressFlaskCloud object storageProcess orchestrator
A concrete velocity proof

What an afternoon builds

Agents logged in a single recent work window — measured straight from the build ledger, not a marketing claim.

0
agents in the window
0
minutes of agent-work
0
tokens processed

A few of the agents in that window

Capability research agentTimeline & archive builderLearning-loop agentVisual extraction agentBuild-economics harvesterStory-page builderFinancials page builderEnrichment wave

Build-ledger subset above is tracked agent tasks only. Across 108 Claude Code sessions: 163M tokens processed. cost is an estimate

We dogfood our own tools

The optimization loop runs on this site too

Our own accessibility and performance checks run against clovitek.ai itself. We show before/after only where we have the receipts.

Before

Unoptimized baseline

Heavier render, unaddressed accessibility gaps — flagged by our own harvester.

After

Optimization KB applied

Verified-safe performance, accessibility, and SEO fixes applied from the KB.

How these numbers are gathered

Sourced, not invented

Every figure on this page comes from a re-runnable metrics collector reading real sources: the agent build ledger (measured per-agent telemetry), a live probe of public product URLs, a repo-code sweep, and live entry counts of the knowledge-base stores. Anything the collector can't measure with confidence is flagged rather than invented.

Provenance tags: Measured = real telemetry or a live count. Estimate = derived (e.g. dollar cost from tokens × a blended rate). Indicative = counts of our own code over named roots only. Same standard used on Our Fleet.

Built in the open

Every number on this page is real. Talk to the founder.

One founder, one shared intelligence layer. The same engine that built this page can build the next one.