The tutor that pushes you to real understanding.

SocsAI turns your code, data, and docs into lessons, probes what you actually know, and coaches with hints before answers, calling concepts back until they stick.

Early cohort invites go out in small batches. No newsletter spam.

Live session · scripted
Look over this code? Retrieval quality tanked after I switched document loaders.
SocsAIFound it — and it traces back to semantic chunking. Remember the training-manual example? What does this splitter do to a procedure that spans two pages?Lesson 04 · Semantic chunking
…it splits mid-procedure, so the two halves embed separately?
Exactly. Next hint: look at what your new loader emits as one “document” — where would you re-join before splitting?
Hint 1 of 4 · Nudge

The full answer is always available — revealing it is just always a choice.

The Socratic method, for people who build.

That's Soc. Socrates asked questions in the agora; Soc asks them in your codebase. He reads your work, remembers what you understand, and refuses to skip the part where you think because that's the part that compounds.

(Yes, those are socks.)

You shipped it. Could you defend it?

Copilot wrote the function. ChatGPT fixed the bug. The pipeline works but somewhere in there, your understanding stopped compounding.

It works until it doesn't. Then the debugging is archaeology, the interview question lands on exactly the thing you pasted, and the judgment you meant to build never showed up.

How it works

One loop, from your material to your own mental model.

01

Bring real material

Upload the notebook, point at the repo, drop in the PDF. SocsAI reads what you're actually working with and teaches on that, not toy examples. Everything it tells you is grounded in your sources or a cited search.

Sources · what the tutor knows about

rag_pipeline.pyindexed
eval_runs.ipynbindexed
github.com/you/rag-serviceindexed
vendor-api-docs.pdfindexed

4 sources read · every lesson claim cites one of them (or a live search).

02

Lessons, generated for your gap

Each lesson is built by a research-first pipeline: outline the concepts, gather evidence from your material and the live web, draft, then review. The writer never sees raw sources, only vetted research, so it can't make things up.

Generating · Lesson 05 — reranking

Outlineconcepts mapped to your goal
Researchyour sources + one live web search
Draftwritten from vetted research only
Reviewchecked against the brief
03

Lessons open by asking, not telling

Every lesson starts with a think-prompt: commit to a position before the explanation. Your answer fuels the system further. What it reveals shapes what the lesson emphasizes next.

Lesson 03 · Retrieval — opening prompt

Think first

Your users ask pointed questions against 400 pages of internal docs. Before reading on — why retrieval, instead of stuffing everything into the context window?

Cheaper per query… and long context dilutes attention?
SocsAIBoth right. There's a third reason involving freshness — watch for it in §2, then tell me which of the three actually applies to your docs.
04

Get interrogated with a ladder, not a wall

In session, the tutor probes why you chose what you chose. Stuck? Hints come graduated: nudge, targeted hint, worked sub-step, full answer. Revealing is always possible — and always a choice you make, not a default you fall into.

Hint ladder · rung 2 of 4

1 · Nudge
2 · Targeted hint
3 · Worked sub-stepnext
4 · Full answeryour call

Rung 4 asks “Sure? You're close.” — then gives you everything, no scolding.

05

Progress that's measured, pitfalls that come back

Every exchange updates a per-concept mastery model. Shaky concepts get flagged, resurface in quizzes, and get called back in later sessions until “I think I get it” becomes “let me show you.”

Your map · 3 of 14 concepts

semantic-chunkingStrong
eval-metricsBuilding
context-mgmtNeeds review

⚑ context-mgmt missed twice — queued for your next quiz and Thursday's session.

Not another chatbot

ChatGPT

gives you code you'll paste and forget.

Courses

teach everyone the same thing in the same order.

SocsAI

teaches you, on your project, until you can explain it back.

Fair questions

Won't it just refuse to help me?

No — the ladder always ends in the full answer. SocsAI just makes revealing it a deliberate step instead of the default, because the choice is where the learning happens.

What topics can it teach?

Anything technical you're trying to actually understand — coding, data science, ML, AI systems. The first cohort goes deep on AI engineering: RAG, context management, orchestration, human-in-the-loop.

What does it cost?

Free during the beta. The early cohort locks founder pricing when paid plans arrive.

Why “Socs”?

Socrates. Also, socks. We know.

The answer is a question away.

Early cohort invites go out in small batches. No newsletter spam.