Skip to main content
/skylab

Senior judgment. Department throughput. One team.

skylab is the agency built for the new era. A senior-led two-person team, modern tooling, and a process that ships at the pace of a department — with every line of code reviewed before it reaches you.

More than a decade of experience helping stealth founders and startups.

/start a project/see how we work

we reply within 24 hours

Software teams got faster. Most got sloppier. We took the speed and kept the bar.

  1. /01YOU

    You bring the problem

    A feature, a bug list, an AI automation, a vague idea. We read your codebase and your KPIs, then send back a written brief inside 24 hours.

    • scope
    • KPIs
    • approach
    • timeline
    • risks
  2. /02WORK

    Wide and deep at the same time

    Several features advance in parallel, or one hard problem gets sustained focus. Tooling scales the throughput — the judgment stays ours.

    • parallel
    • our judgment
  3. /03US

    Every output flows through us

    Nothing ships without a human reading it. A senior engineer reads the diff line by line.

    • every PR
    • every commit
    • every deploy
  4. /04YOU

    You get human-paced delivery

    Commits in chunks you can actually review. Daily async updates. One contact, one team.

    • daily updates
    • one contact

Nothing reaches you that has not been reviewed by a human.

The AI never talks to the client. That separation of concerns is not a limitation — it is the whole point.

Recent projects.

run console / t-417stopped at a ready pull requestrunning
task

T-417

Let admins rename saved filters

marked by
teammate
worktree
wt/t-417 isolated
order
tests → code
risk → blind checkers
U1lowU2medU3high
acceptance criteria
  • An admin can rename a saved filter
  • Two filters cannot share a name
  • The new name shows at once
run trace
ready
teammate marked it ready
write tests
FIRST
write code
U1U2U3in wt/t-417
checker a
U1U2U3
checker b
U2U3
checker c
U3
types
lint
tests
criteria

A person
decides.

no merge
step

inspectPick a lane or a unit.

09:12:04readya teammate marked T-417 ready09:13:52testswritten first, before any code09:41:30checksU3 risk high → 3 blind checkers09:58:17stopready pull request · no merge step

Illustrative mockup.

AI engineering· In production

AI delivery pipeline

A delivery pipeline we designed and built for a client’s engineering team, in production on a large TypeScript codebase. Once a person marks a task ready, it writes the tests first, then the code, in an isolated workspace — and one to three blind AI checkers, more for riskier work, review each unit before types, lint, tests and acceptance criteria run. It stops at a ready pull request: there is no merge step, and a person decides.

blind AI checkers, scaled to risk
1–3
automated gates: types, lint, tests, criteria
4
merge steps — a person decides
0
/request project information
review queue3 waiting for a personqueue clear
confidence
threshold
0.700.800.90
auto-tagged9to a person3
Item A1, auto-tagged, 0.97Item A2, auto-tagged, 0.91Item A3, needs a person, 0.74Item A4, auto-tagged, 0.94Item A5, auto-tagged, 0.86Item A6, auto-tagged, 0.93Item B1, needs a person, 0.58Item B2, auto-tagged, 0.89Item B3, auto-tagged, 0.95Item B4, auto-tagged, 0.84Item B5, needs a person, 0.66Item B6, auto-tagged, 0.92
human review gate
item B1needs a personreviewedauto-tagged

Is the fill striped?

vision model
shape
square
0.95
fill
striped
0.58
marks
two
0.91

Below 0.80 on any attribute, a person decides.

solidstripeddotted

Illustrative mockup.

E-commerce + AI· Enterprise

Wellness e-commerce + AI ops

Built a production-scale pipeline for influencer-driven e-commerce with thousands of products that needed intelligent metadata and categorization. Deployed a custom vision LLM pipeline that classifies content at scale, with human review gates on every low-confidence result.

items classified
7.8k+
accuracy
96%
lower cost than manual tagging
40%
/request project information
Mockup view

learning modules

Low endDynamicsVocals

+ more modules

answerVocalsAI behaviour v13

you asked

Every S on my vocal sounds harsh. How do I tame it without dulling the take?

Tame the S’s, not the whole vocal.

  1. Find the band

    Sweep a narrow EQ boost from 5 to 9 kHz.

  2. Duck only that band

    Split-band de-esser, 3 to 6 dB of reduction.

  3. Stop before the lisp

    If an S turns into a “th”, back it off.

try these settings

freq
6.8kHz
range
−4dB
mode
split
answeradaptive quizai behaviour

question 4

Which band usually holds harsh vocal sibilance?

A 200–400 HzB 1–2 kHzC 5–9 kHzD 15–18 kHz

correct

The next question is harder.

difficulty3 → 4

versions

no deploy
  1. v14draftEditor
  2. v13activeCoach
  3. v12Producer
  4. v11Editor

v13 → v14 · instructions

Answer like a senior engineer.

List every option you know.

Lead with one fix, then say why.

Ask for the genre when it matters.

this drill

source
noise
boost
+6 dB
bands
octave
rounds
10

how a round works

  1. 01listen
  2. 02pick a band
  3. 03see the curve
Noisefrequency drill06 / 10

noise · +6 dB · octave bands · 10 rounds

answer

Boosted at 2 kHz

your pick 2 kHz· right

63 Hz125 Hz250 Hz500 Hz1 kHz2 kHz4 kHz8 kHz16 kHz
0:02 / 0:02

Illustrative mockup.

took over a prototype · led the paid launch · later retired itmaintained and shipped

Specialized APIs· Indie → growth

Audio production tooling

Built a suite of music and audio tools: deep integrations with streaming APIs, generative audio testing for creator workflows, and a dedicated sales surface. Three repositories, one team, one vision. Domain-specific engineering in a vertical most agencies cannot touch.

repos shipped
3
APIs integrated
4+
team, end to end
1
/request project information
song app

New song

Write a message, then pick a style.

message
Happy birthday, sis! Dinner at eight, and the cake is on me
59/200
song style

HappyPartyPop

make my song

Song style

Blend a vibe with a genre.

vibesup to 2

ChillHappyPartyUpbeatDreamyCalmMoodyNostalgic

genre

PopAcousticHip-hopRockLo-fiCountry
use this style

PreviewMaking your song…

Your song is ready

take 10:12

Happy · Party · Pop

Happy birthday, sis! Dinner at eight, and the cake is on me

new take

Not quite right? Try a new take.

Illustrative mockup.

Mobile + AI· Startup → live

AI-powered consumer music app

An iOS app that turns text messages into AI-generated songs. Multiple music generation API integrations, real-time audio processing, lyrics synchronization, StoreKit subscriptions, A/B testing, and analytics. Live on the App Store with real paying users. From concept to shipped product in weeks.

AI APIs integrated
3+
on the App Store
live
team, end to end
1
/request project information
control plane
S-7F3AS-21C9S-B04ES-6D1F
sitelive

live00:12:4148.2 MB

isolated remote browser

capturedrebuiltcode stripped

isolation
container
own
identity
own
user IP
not sent
user cookies
not sent
ssrf guard
public siteallowedprivate serverblockedinternal serverblocked
egress fleet

health checksauto-recovery

meter

per session00:12:41

per byte48.2MB

bytes

Illustrative mockup.

Browser infrastructure· End to end

Remote browser infrastructure

We built and ran the server side of a product that opens any website in a remote browser. Every session gets its own container and identity, so the user’s IP and cookies never reach the site; each page is captured and rebuilt with anything that could run code stripped out, and no session can reach a private or internal server. A self-healing egress fleet keeps it up, and usage is metered per session and per byte.

container and identity per session
1
user IPs or cookies passed to the site
0
usage metering, per session
per byte
/request project information
sync console / eu retailtwo-way · real-time
storefrontonline
item 01, 3 in stockitem 02, 1 in stockitem 03, 5 in stockitem 04, 2 in stock
checkoutitem 03

ordercompliant paymentpaid

sync

ordersstock

purpose-built0 plugins

accounting systemlegacy

itemon hand

item 01on hand3

item 02on hand1

item 03on hand5

item 04on hand2

ledger

in storesaleitem 04−1

in storesaleitem 01−1

onlineorderitem 03−1

Illustrative mockup.

Integrated commerce· Custom

Retail commerce + POS sync

Custom e-commerce storefront for a European retail market with real-time inventory sync from a legacy accounting system. Two-way communication, compliant payments, purpose-built infrastructure. Not a template, not a plugin — the right thing for a specific market.

real-time sync
2-way
compliance
EU
plugins required
0
/request project information
ops hub / agendalive
platforms
  • platform a2
  • platform b2
  • platform c1
  • platform d1

1 queue

one queueauto-prioritized
T-21, Fix a broken footer link, in reviewT-34, Rename a settings label, to doT-12, Add alt text to icons, in reviewT-47, Write tests for sign-in, to doT-08, Tidy spacing on mobile, to doT-16, Update a dependency, live

+1 more

ai briefing

Six tasks from four platforms. Two wait for review, the footer link fix first. The dependency update is live.

task
T-21in review

Fix a broken footer link

client
client a
from
platform b
priority
3 of 3 marks

Waiting for a person.

Illustrative mockup.

Agency infrastructure· Product

Internal operations hub

The multi-client operations platform we built for ourselves. Task aggregation, live agenda dashboards, AI briefing, human review, deployment orchestration. We eat our own cooking — every process we learned from clients, we automated for ourselves. This is what you get when you hire us.

platforms aggregated
4
queue, auto-prioritized
1
live dashboards
24/7
/request project information
product stack / the whole chainlive in production
every layer
Frontendwhat users touchowned by the teamBackendlogic and dataowned by the teamInfrastructurewhere it runsowned by the teamMonitoringhow it is watchedowned by the team
inspect

Frontend

What users see and touch. Novel UX patterns.

owned by
the team
status
live

last request

frontendbackendinfrastructureresponse

The team owns every layer.

Illustrative mockup.

Full product engineering· Startup

Consumer product build

End-to-end consumer product from vision to live application. Frontend, backend, infrastructure, monitoring, the whole chain. Novel UX patterns. Careful performance tuning. Live today, users engaging, the team owns every layer. We focus on products that stay live, not features that get abandoned.

stack coverage
100%
in production
live
by the team
owned
/request project information
/start a project

Privacy-first, by default.

We never publish who we work with. No client names, logos, screenshots or data — every engagement runs under NDA.

Want to know more about a project? Ask us. With the client’s permission, we’ll share more — directly and in confidence.

/request project information

Full-stack. One team. Same standard.

Full-stack web

Next.js, React, Node, Python. Postgres, Supabase, Turso. Real apps with real users in production — e-commerce, SaaS dashboards, content platforms, internal tools. Production standards from day one.

Mobile — iOS + cross-platform

Native Swift for iOS, React Native and Expo for cross-platform. Standalone apps or native layers grafted onto existing web platforms. App Store, Play Store, TestFlight — we handle the whole chain.

DevOps + infrastructure

AWS, Cloudflare, Vercel, NixOS, PM2. CI/CD on every commit, observability that catches problems before users do, zero-downtime deploys and migrations. Deployment is built in from the first commit.

Architecture

Systems sized to the users you have, not a theoretical peak. Multi-tenant by design, event-driven where it matters, AI-native where it genuinely helps users — not for the pitch deck.

AI integration

Claude, OpenAI, Gemini, Perplexity. RAG on private data, autonomous agents, evaluation frameworks that measure output quality. Production LLMs in our stack from the start — we know the seams where they fail and the patterns that work.

Custom solutions

When no existing tool fits, we build the right thing from scratch. Internal dashboards that become an advantage, integrations your competitors do not touch, tools built for one team only — and better for it.

/start a project

What we believe.

  1. /01

    Code is communication.

    It speaks to the next engineer who reads it, not just the machine. We write for humans first. Clarity over cleverness.

  2. /02

    AI is a tool, not a shortcut.

    We use AI only where it is reliable; everything else stays deterministic. The model writes code for a structure already decided — it does not decide the structure. A human always reviews. A human always decides.

  3. /03

    Process should be invisible.

    You should not see the machinery — just one team, one channel, one interface. The complexity stays on our side of the wall.

  4. /04

    Scale without compromise.

    More code, more features, more load should never mean lower quality. We scale the team, not the standard — senior-engineer judgment on every commit.

  5. /05

    Your constraints are features.

    A small team, a tight budget, legacy code, weird infrastructure — not obstacles. They shape the solution. The best systems are the constrained ones.

  6. /06

    Done means live.

    We do not hand off and vanish. We deploy it, monitor it, and respond when it breaks. While we are engaged, the product is not done until it stays live.

FAQ

  • How is this different from hiring a freelancer?

    A freelancer is one person working on a task. We are a senior engineer and a technical project manager backed by a fleet of AI agents, working on an outcome. A senior engineer sits between every output and you. You get the judgment of a CTO with the throughput of a department. Also: no context loss. Our tooling is built for continuity across months, not days.

  • What is the minimum engagement?

    We start with a short diagnostic sprint — typically one to two weeks. We read your codebase, produce a written architectural assessment, and deliver real working code. If we are a fit, the engagement begins on a monthly retainer. If not, you keep everything we delivered. There is no long-term contract lock-in.

  • How do you avoid AI hallucinations shipping to production?

    Humans review before anything reaches you. Every PR, every commit, every decision — a senior engineer reads the diff line by line. The AI drafts fast. The human catches the hallucinations, the missing edge cases, the subtle shortcuts. The AI never talks to you. You only ever see the output we approved.

  • Do you work with existing codebases or only greenfield?

    Both. We have done platform migrations, legacy refactors, feature grafts onto existing systems, and greenfield builds. Your codebase conventions become our conventions. By week one, our output looks like it came from your existing team.

  • Can you work alongside my existing engineering team?

    Yes. We can be your entire engineering team, or we can augment an existing one. If you have senior engineers already, we pair with them — your team reviews our code, our team reviews theirs. The bar stays high either way.

  • What happens if I need to pause or stop?

    You can scale down or pause at any time. End of month, you step back. Code is always yours, always committed, always deployable. If we are a fit, great. If you need to pause, that is fine too. No penalties, no lock-in.

  • How do you measure quality without daily meetings?

    Shipping velocity, code review metrics, uptime, and user feedback. The product either works or it does not. You get access to our live agenda dashboard: every PR, every commit, every deployment, visible in real time. You can review on your own calendar.

  • What if the work breaks something in production?

    We fix it. If we shipped it, we own it. Our bar is code that stays live — if it does not, it is not done. Every PR ships with a rollback plan. One message and we roll it back. We design for reversibility from the first commit.

  • Where are you based? How do time zones work?

    We are based in Lisbon and São Paulo and work with clients across US and EU time zones. Async-first communication means you never wait for our morning. Updates land before your day starts.

  • Who owns the code?

    You do. All code is yours — always committed to your repositories, on your infrastructure. IP is assigned on contract signature. We sign NDAs as standard, and we agree on data-handling terms — including AI tooling policies — before we touch anything.

Have something worth shipping? Let's talk.

PRIMARY

[email protected]

Write with the shape of the project, the deadline, and any constraints. We read and reply to everything within 24 hours.

/[email protected]