New · Multimodal evaluation now in beta

Turn the chaos of free tutorials into a verified path to your next career.

Cairn uses Gemma 4 to plan, verify, and showcase your learning — in 12 weeks. No more bookmarks-as-progress.

Start free
No credit card Free Gemma routing Recruiter-shareable
SCROLL
Trusted by learners from
Stanford
MIT
Berkeley
CMU
Toronto
Princeton
ETH
Imperial
UCL
NUS
Tsinghua
EPFL
Stanford
MIT
Berkeley
CMU
Toronto
Princeton
ETH
Imperial
UCL
NUS
Tsinghua
EPFL
The problem with self-directed learning

You bookmark 47 tutorials.

47 tabs open

You finish 3.

3 completed

You can't prove any of them.

0 credentials
Until now
How Cairn works

Four steps from chaos to credential.

STEP 01
> goal
6 moremoteno degree

Tell us your goal

Free-form, plain English. Gemma 4 4B parses into a structured profile.

STEP 02
Phase 1
Phase 2
Phase 3

Get a 12-week path

A drawn-from-scratch roadmap with phases, milestones, and curated free resources.

STEP 03
import torch
class Model(nn.Module):
def forward(self, x):
return self.layers(x)

Ship → AI evaluates

Multimodal: 27B reviews code, 12B sees screenshots. Stage-by-stage scoring.

STEP 04
CAIRN VERIFIED
AI Engineer

Earn verified credentials

HMAC-signed badges on a recruiter-shareable portfolio. No more 'trust me bro.'

The hero feature

Multimodal evaluation, in four stages.

We don't just run your code — we look at the screens you ship. Gemma 4 12B's vision sees what users see.

Stage 2 + 3 · Live preview
codex-studio / src / agent.py+72 −18
# Stream multimodal eval
async def evaluate(repo, shots):
graph = build_graph(repo)
review = await gemma27b(graph)
vision = await gemma12b(shots)
return sign(review, vision)
+ structured findings: 14
+ verified: True
app.codex.studioVision · 12B
codex-studio
What do you want to build?
headerh1cta cluster
Stage 1 · Structural
Commits
0
Tests passing
0%
README quality
0.0/10
Stage 4 · Credential
CAIRN VERIFIED
#0x1BB78
AI Engineer
Codex Studio · Apr 2026
SCORE
92/100
sig: e3b0c4 ... 9b427
Provider fallback chain

If a provider is rate-limited or down, we fail over without losing your run state.

Gemma 4 27BGemini 2.5DeepSeek R3Local 4B
· latency p50 1.4s · success 99.7%
Public portfolio

One link. Real proof.

Every credential is HMAC-signed and verifiable. Recruiters click, scan a QR, and see the underlying evaluation.

cairn.dev/u/mira-k
secure
Mira K.
AI Engineer · Bangalore
Cairn-verified path
AI Engineer · Week 12 of 12
Codex Studio
92/100
VisionRAG
NoteShelf
88/100
Next.jstRPC
Echo Buddy
81/100
WhisperEleven
PaperCanvas
95/100
Three.jsGLSL
Intentional model selection

Why Gemma 4, three sizes deep.

Different problems want different brains. We route each task to the smallest model that handles it well.

Gemma 4
4B
latency p50 · 120ms
Goal parsing

Tiny, fast, runs everywhere. Converts your messy goal into a structured profile.

Gemma 4
12B
latency p50 · 480ms
Visual review

Vision-tuned. Scans your screenshots and reads UI like a senior designer.

Gemma 4
27B
latency p50 · 1.4s
Code review

Deep reasoning. Reads your repo as a structured graph, not just files.

FAQ

The honest answers.

Still curious? Email hi@cairn.dev — we read everything.

Cairn never sells you content. We curate the best free resources, pace them, and — most importantly — verify what you shipped. The credential lives outside Cairn: it's an HMAC-signed JSON anyone can check.

For each project, we run four stages: structural (commit graph, tests, README), code review (Gemma 4 27B reads the AST), visual review (Gemma 4 12B looks at screenshots of your running app), and synthesis (signed credential if score ≥ 0.65).

Every task has a fallback chain. The default is Gemma 4 27B → Gemini 2.5 → DeepSeek R3 → Local 4B. You can customise this per-task in admin.

Yes. The badge contains a signed JSON payload. Recruiters can verify it against our public key or run their own verifier — no Cairn account required.

It works best for technical paths: software engineering, ML, design engineering, devrel. For very niche or non-tech careers we ship a 'best-effort' path with a warning.

No. We pull the repo at evaluation time, build a structural representation, and discard the source. The evaluation report is yours.