Content Ingestion for AI-Native Learning — SQL Walkthrough

Bounded prototype walkthrough showing how partial SQL teaching material can be transformed into a structured, inspectable learning graph with critique, repair, clarification, and deterministic validation.

Date:

Role: Prototype / system-design walkthrough

Stack: Python, structured learning graphs, deterministic validation, SME clarification loop, static HTML artifact

Outcomes

  • Shows a bounded SQL ingestion slice from draft graph through final interpretation
  • Makes critique, auto-repair, and clarification inspectable rather than hidden in prompts
  • Establishes a trust boundary: confirming an edge is not the same as proving node provenance

Prototype artifact

This is a bounded walkthrough of a content-ingestion system for an AI-native learning platform. The example is deliberately scoped to intro SQL on a single-table slice so the workflow, trust boundaries, and validation logic stay inspectable in an interview setting.

The core problem is not file upload or chunk-and-embed. It is transforming messy SME material and legacy learning assets into structured learning ingredients that can be reviewed, repaired, and used safely by downstream adaptive systems.

Why this artifact exists

The walkthrough makes a concrete claim about system shape: start from partial source material, build a draft learning graph, expose uncertainty, recover bounded facts through clarification, and keep deterministic validation in charge of what is allowed into the final graph.

This page presents the artifact as a portfolio and system-design example, not as a claim that this exact pipeline was shipped unchanged in a prior employment setting. The value here is the architecture, inspectability, and the way the trust boundary is made explicit.

What this artifact shows

Draft graph

An initial structured graph assembled from partial SQL teaching material, including skills, objectives, assets, and prerequisite candidates.

Critique

A review pass that turns hidden ambiguity into explicit structural and support issues.

Auto-repair

Low-risk fixes that improve honesty and completeness without inventing new truth.

Clarification loop

A bounded question to an SME-like source, replayed across different answer styles.

Deterministic validation

A strict gate that accepts only schema-valid, in-scope updates and rejects plausible but loose answers.

Final interpretation

A final graph reading that separates scoped edge confirmation from broader provenance claims.

Key system choices

1

Skeleton first, assets second

The walkthrough starts by making learning structure explicit before trying to perfect downstream teaching assets.

2

Deterministic integrity, model-assisted semantics

Models help interpret and compress ambiguity, but deterministic validation remains authoritative on what changes are allowed.

3

Sparse human judgment

The human loop is kept narrow and high-value: clarify a bounded relation instead of asking for wholesale graph approval.

4

Edge confirmation is not node provenance

The running example makes the trust boundary legible: confirming sql_order_by -> sql_limit does not prove full source support for sql_limit.

5

Bounded evaluation and inspectability

Every stage is designed to be reviewable enough to discuss failure modes, trust modes, and operational release decisions.

Why it matters for AI-native learning

For adaptive learning products, the hard part is often not generation quality in isolation. It is whether the system can convert inconsistent teaching material into structure that is reliable enough for sequencing, remediation, and evaluation.

This artifact is relevant because it shows how I think about structured AI systems, human-in-the-loop transformation, evaluation, and trust boundaries. It is also a useful production-style prototype for discussing where semantics can be model-assisted and where integrity needs deterministic controls.

Walkthrough artifact

Full walkthrough

The hosted artifact is a self-contained HTML walkthrough from the SQL content-ingestion prototype repository. It is best read as a bounded design and evaluation artifact.

Open Full Walkthrough Open Reading Guide View Source Repository

Scope: single-table intro SQL slice, draft graph to final interpretation, with a focused clarification example around sql_limit and sql_order_by.

If the embedded preview feels cramped, open the full walkthrough in its own page. The reading guide is a faster route for interview prep.