archive · case_studies · 12 verified builds

Real systems. Real numbers.
Receipts attached.

Q: How does kratt measure case study results?

Every case study reports the metric the client asked us to move, call response time, consultations booked, hours returned, revenue lifted, ad spend saved, measured against a baseline agreed in writing before the engagement starts. No vanity metrics, no cherry-picked windows.

Q: Will kratt share my case study publicly?

Only with written approval. Public case studies are reviewed line-by-line with the client and never go live without sign-off. Many kratt clients prefer to keep wins private, that is fine, we have a parallel library of confidential case studies shared only on strategy calls under NDA.

Q: Can I talk to a kratt reference client?

Yes, after the second strategy call. We introduce qualified prospects to a current or recent client in a similar vertical. References take a 20-minute call directly with you, no kratt operator on the line, no script, no filter.

Every build below is in production. Every metric was measured against a baseline we agreed on in writing before we started. Real clients, real tech, real ROI ranges from 260% to 620%. Pick a story below and see exactly how the loop closed.

aggregate · production · today

platforms shipped

avg ROI

hours reclaimed/wk

avg time-to-ship

~0w

[ featured_builds

Four flagship builds. Full receipts.

Each story includes situation, challenge, what we shipped, the architecture, the numbers, and a quote from the operator. No fluff, no rounded marketing-deck numbers.

build_snapshot · rahamaa-full-cycle Live · ongoing iteration

Hermes outreach harness

Lead qualification (Puppeteer + Gemini vision)

Auto-generated landing site per prospect

Per-pipeline ElevenLabs voice agent

Pipedrive + Airtable + autopilot dashboard

Hermes outreach harnessNode.js orchestratorClaude (primary)DeepSeek + Kimi (fallback)Gemini visionImagen 4PuppeteerElevenLabs (3 agents)TwilioPipedriveAirtablen8nVercel BlobNext.js (website-gen.net)

01 / Outbound sales automation · External · Estonia / Finland Outbound sales, on autopilot

Three pipelines, one autopilot, outbound sales running without an ops team

Hermes harness pulls leads. Every prospect gets a personalised landing page in minutes. Voice agents pick up live across three pipelines. Every agent run logged to the dashboard.

AI Implementation External sales-automation suite Reference on request

situation

The client needed to run outbound sales across three distinct pipelines (EE finance, FI market, dental) simultaneously. Manual outreach couldn't scale: every prospect required a tailored landing page, a qualification pass, a localised cold-call, and a CRM update, multiplied across three verticals.

challenge

Build one orchestrator that runs all three pipelines on autopilot. Per-prospect personalisation (not template spam). Per-pipeline voice agent with the right language + script. Every agent run captured so the dashboard shows real numbers, not theatre.

what we shipped

A full pipeline: the Hermes harness ingests leads, Puppeteer + Gemini vision qualifies each website, the Node.js website-gen orchestrator builds a personalised React landing page per prospect (Claude as primary writer, DeepSeek + Kimi fallback for resilience, Imagen 4 for hero assets) and deploys it to a website-gen.net subdomain. Inbound voice routes to the matching ElevenLabs agent (EE cold-caller, FI cold-caller, dental cold-caller). Pipedrive holds the deal state, Airtable mirrors lead status, every agent run posts to the autopilot endpoint so the dashboard sees what actually shipped.

what didn’t work, at first

When all three pipelines ran in parallel, the primary Claude calls rate-limited and broke landing-page generation mid-batch. We added DeepSeek + Kimi as automatic fallbacks and a queue retry layer in week 2, orchestrator hasn't dropped a build since.

pipelines live

Per-prospect

auto-generated LP

720ms

voice pickup

100%

agent runs logged

"Hermes pulls the leads, the orchestrator builds the page, the agent makes the call, the dashboard logs the run. Three pipelines, one operator at the top."

kratt-built engagement

build_snapshot · marketing-intel 6 weeks

7 platform APIs

Make.com (daily)

Airtable SSOT

React dashboard

PDF + Slack

Make.comAirtableReactGA4 APIMeta Graph APIYouTube API

02 / Real Estate · Bali 380%

7 marketing platforms unified, 3 hours/day saved

A vertically integrated Bali developer drowned in cross-platform reporting. We replaced Monday-morning spreadsheets with a live dashboard.

AI Implementation Vertically Integrated Real Estate Developer · Bali Reference on request

situation

Marketing data scattered across Google Analytics, Search Console, Instagram, Facebook, LinkedIn, YouTube, Mailchimp. The team spent 15+ hours weekly pulling numbers manually. By the time reports compiled, the data was stale.

challenge

Cross-platform attribution was impossible. Decisions were being made on month-old data while ad budgets bled.

what we shipped

Make.com scenarios scrape daily metrics from all 7 platforms into a centralized Airtable. A custom React dashboard pulls via API with cross-platform comparison, trend analysis, one-click PDF reports.

what didn’t work, at first

Three of the seven APIs (LinkedIn, YouTube, Mailchimp) ratelimit aggressively. We added a token bucket + retry layer in week 3 after the first sync hit the wall.

7→1

sources unified

$96k

annual labor saved

3h/day

reclaimed

100%

data accuracy

"We used to spend 3 hours every Monday pulling numbers from 7 platforms. Now the dashboard updates itself daily. Cross-platform comparison alone changed how we allocate ad budget."

Head of Marketing · Bali Real Estate Developer

build_snapshot · propertybase 21 days

Subdomain router

Supabase RLS

Tenant data

Role-based UI

Per-tenant dashboard

React 18TypeScriptSupabaseTailwindEdge FunctionsVercel

03 / Proptech SaaS · diil.ai Shipped in 3 weeks vs 4-6 months

Multi-tenant proptech SaaS, shipped in 21 days

Quoted $80–150K and 4–6 months by traditional agencies. We shipped it in three weeks for a fraction.

VibeCode diil.ai, proptech SaaS for property developers See it live · diil.ai

situation

diil.ai needed a centralized proptech platform where each property developer tenant could manage listings, track leads, and report to investors, on their own subdomain. Agency quotes for the same scope came in at $80–150K with 4–6 month timelines.

challenge

Three new tenants launching simultaneously. The team needed working software now, not in six months.

what we shipped

VibeCode rapid build: a multi-tenant proptech SaaS in 21 days. Subdomain-based tenant isolation, Row Level Security, RBAC, listings, leads, investor reporting. Phase 12 (Commission Binding + Pickers + Multi-Buyer) shipped April 2026.

what didn’t work, at first

Subdomain tenant isolation with Supabase RLS was the hard part: an early version leaked one tenant's view of pipeline stages into another. Caught in QA, fixed before launch.

21d

concept → prod

$38k

project cost

tenants live

200h

eng saved

"I was skeptical when they said three weeks. Quoted 4-6 months by traditional agencies. They shipped a working multi-tenant SaaS that our tenants actually use daily."

Team, diil.ai

build_snapshot · lead-engine 12 days

Apify discovery

Gemini scoring

Personalised LP

Cloudflare Worker

Sales Telegram

n8nApifyCloudflare WorkersGemini 2.0Google SheetsTelegram Bot

04 / Agency · Programmatic 560%

A programmatic sales machine, 200+ qualified leads/month

Manual prospecting capped them at 20–30 conversations/month. We built the engine that finds, scores, and personalizes, for $0.50 a lead.

VibeCode Digital Agency · Pipeline build-out kratt-built · pipeline for our own outreach

situation

The agency relied on referrals + manual cold outreach. Sales spent hours per prospect. Pipeline was capped at 20–30 conversations/month, well short of growth targets.

challenge

No scalable system for lead generation. Manual research could not keep up with growth.

what we shipped

A full programmatic lead engine: Apify discovers businesses, Gemini 2.0 Flash scores website quality, n8n generates a personalized landing page per prospect, Cloudflare Workers serves it, Telegram bot pushes hot leads to sales.

what didn’t work, at first

Gemini scoring drifted on edge categories (single-page brochure sites with strong copy). We added a human-in-the-loop review queue for low-confidence scores before they pinged sales.

12d

system built

200+

qual leads/mo

<$0.50

cost / lead

manual research

"They built us a system that finds businesses, scores their websites with AI, generates personalized landing pages, and delivers ready-to-call leads. Pipeline went from 20 conversations a month to over 200."

Founder · Digital Agency

[ archive_grid

Eight more. Same rigor.

Quick-look cards for the rest of the recent archive, AI implementations, profit audits, and rapid VibeCode builds. Each one shipped, each one paid back.

AI Implementation Property Development · Bali $200 → $6 per render

AI render pipeline cut style variations 85%

45m→3m

per render

85%

cost cut

Went from $200/render with external 3D artists to 3 luxury variations auto-generated. Moroccan and Italian variants outsell originals.

n8nFreepik APIGeminiDrive 3 weeks

Reference on request

AI Implementation Real Estate Investment · Bali Sales-cycle compression

Contract turnaround: 2 days → 15 minutes

2d→15m

turnaround

$54k

annual save

Sales clicks a button in Airtable, fully populated PPJB lands in the buyer inbox in 15 minutes. 3 deals closed they would have lost on the old timeline.

n8nPandaDocAirtableSlack 4 weeks

Reference on request

AI Implementation AI SaaS · diil.ai Compounding organic traffic

100+ SEO posts/month, publishing on autopilot

100+

posts / mo

human hours

Python engine pulls keyword clusters, generates long-form posts with brand-voice prompts, ships hero images via DALL-E, publishes to a Supabase-backed /blog. The engine runs on a cron, no one touches it.

PythonSupabaseClaude + OpenAIDALL-E/blog CMS 3 weeks

See it live · diil.ai/blog

VibeCode Real Estate Fund · Bali New capital committed

Investor portal in 14 days, 3 new LPs attracted

14d

shipped

new LPs

Replaced month-old PDF reports with a live dashboard. Real-time portfolio + project tracking + branded PDF export. 3 LPs cited the portal as why they committed.

ReactAirtable APIRechartsjsPDF 14 days

Reference on request

VibeCode Towing + Transport · Estonia Local-pack visibility

Estonian puksiir site, Lighthouse 99 / 100 / 100 / 100

99/100

Lighthouse perf

95/A

SEOmator score

amobile.ee rebuild: structured data for LocalBusiness + Service, Estonian copy, instant-load on every device. Now the canonical Tartu reference for puksiir queries.

Vite + ReactTailwindSchema.orgVercel Shipped 2026-05-19

See it live · amobile.ee

AI Implementation DTC Content Ops · kratt-built Hundreds of creatives / month

A Python pipeline that ships DTC creative at scale

Live

production pipeline

Per-brand

codified DNA

Takes a codified brand DNA file and outputs ads, social, video scripts, outreach assets on a pipeline. Batch generators + prospect samples + webhook server, the whole production line runs without a designer touching it most days.

PythonBrand DNA specImage genWebhook server Internal R&D · ongoing

kratt-built · the engine behind Ad Factory

Voice Agent Real Estate · Bali Founder time, recovered

Estonian voice agent books warm discovery calls 24/7

720ms

pickup latency

EE → EN

mid-call switch

Estonian-speaking voice agent on a +372 number for inbound property buyers. Qualifies in Estonian using the ERALDA framework, books discovery slots straight into the founder calendar. Twilio + ElevenLabs + n8n webhooks to Cal.com and Airtable.

Claude Haiku 4.5ElevenLabsTwilio +372n8nCal.comAirtable ~45 min · deploy package

Reference on request · Investland Bali

AI Implementation Electrician · Tartu, Estonia Inbound from voice + local

Estonian electrician brand built for AEO + GBP local pack

service spokes

AEO

voice-search ready

GBP-anchored 6-spoke site for an Estonian electrician brand. Pricing calculator on every spoke, FAQPage schema for AEO, directory submissions, all wired to capture voice-search + "near me" local-pack queries.

Vite + ReactSchema.org FAQPageGBPPricing calculator In delivery

Reference on request

[ live_screens

What it actually looks like.

Six recent ship-and-deploy projects with screenshots from production. Hover any tile for the brief.

Live

Battleground

Sports · Esports

Live

Gena

AI · Estonia

Live

Kaevetehnik

Trades · Estonia

Live

Sinu Bistroo

Hospitality · F&B

Live

Nitro Consulting

B2B · Consulting

Live

Caldoring

D2C · Wellness

[ partner_platforms

Platforms we operate.

Long-term partners and our own products. Different stack, different sector, same backbone, ours to ship, yours to own.

Live

Live

PlatformContent EngineLead Gen

Live

Live

Construction · Indonesia

Live

Live

Live

Live

[ services_tally

What we ship, by the count.

Across every build above and the work running in the background. If we offer a service somewhere on this site, the number below is how many times we've actually shipped it in production.

production sites

Custom Next.js / React. Real code, your repo. Lighthouse 95+, schema-clean, A11y-conscious.

0×

voice agents live

Sub-second pickup, brand-trained, multi-language, calendar + PMS integrated. Running 24/7 in production.

automations running

Make.com, n8n, native API, custom code. Lead routing, contract gen, billing, ops, reporting. All monitored.

0×

SEO programs

Technical foundation + AI content engine + human edits. Page-1 rankings within 90d. 12 posts/mo on autopilot.

0×

profit audits delivered

Construction, e-com, property mgmt. Audits found $42k–$180k of waste each. ROI inside 60 days.

0×

multi-tenant SaaS

PropertyBase + others. Subdomain isolation, RLS, RBAC, role-based dashboards. Shipped in 21 days each.

[ next_build

Want to be case study #15?

30-minute call. Screen on. We map your stack live, find the 3 highest-ROI loops, and tell you what they'd cost to ship. The audit is free.

[ book_audit ]

Case Studies · FAQ

How we report results, and how to vet us.

The metric we use, the consent rules around publishing, and how to talk to a reference client. Three answers most procurement teams ask before they buy.

How does kratt measure case study results?+

Every case study reports the metric the client asked us to move, call response time, consultations booked, hours returned, revenue lifted, ad spend saved, measured against a baseline agreed in writing before the engagement starts. No vanity metrics, no cherry-picked windows.

Will kratt share my case study publicly?+

Only with written approval. Public case studies are reviewed line-by-line with the client and never go live without sign-off. Many kratt clients prefer to keep wins private, that is fine, we have a parallel library of confidential case studies shared only on strategy calls under NDA.

Can I talk to a kratt reference client?+

Yes, after the second strategy call. We introduce qualified prospects to a current or recent client in a similar vertical. References take a 20-minute call directly with you, no kratt operator on the line, no script, no filter.