Learning pathsPricingChecking sign-in…
← ML curriculum

AISD-001 · DESIGN PRACTICE

Practice: AI coding assistant with repository context and safe tool execution

Draft the complete decision system before revealing the guidance. Use a 45-minute rehearsal budget and adapt it to the interviewer’s direction.

Preserve eligibility, available-at-decision inputs, mature labels, bounded work and compatible releases. Explain alternatives and test a normal, boundary and failure case.

Your practice draft

Design an AI coding assistant for a multi-tenant developer platform. It must answer repository questions and propose safe changes across large repositories. Explain indexing, retrieval, context provenance, patching, sandboxing, approval, recovery, evaluation and capacity.

Saved locally on this device. The worked answer does not score your draft. Existing lesson notes and completion are separate.

Progressive hints

Hint 1

Start by pinning every task to an authenticated principal, repository and immutable base commit.

Hint 2

Separate repository indexing and online retrieval; combine symbol/lexical/semantic search and verify blob SHAs.

Hint 3

Treat model output as an untrusted patch proposal. Use an isolated sandbox and deterministic policy broker.

Hint 4

Bind approval to the exact patch hash, reauthorize, and compare-and-swap the branch head before writing.

Self-review evidence

For each criterion, record Missing, Prompted, Independent or Adaptive reasoning. No automated scoring or offer prediction is provided.

  • Identity, tenant, repository and commit scope are explicit.
  • Index freshness, webhook reconciliation and exact source provenance are addressed.
  • Prompt injection, secrets and untrusted repository code are handled.
  • Patch paths/blob SHAs are validated and writes are approval-gated.
  • Sandbox has resource and network isolation with no ambient credentials.
  • Retries, cancellation, idempotency and branch conflicts are recoverable.
  • Quality, safety, latency and cost are measured separately.
Reveal worked design guidance

A strong design uses a provider connector with short-lived scoped capabilities; an event-driven indexer publishes immutable commit snapshots; a retrieval service combines symbol, lexical and dense search; a context builder records exact source spans; an LLM proposes a patch; a policy broker validates it; an ephemeral sandbox runs tests and scanners; the user approves the exact diff; and a write broker reauthorizes and uses compare-and-swap against the branch head. Durable task state and idempotency make retries safe. Read-only explanation remains available when write policy or sandbox execution is unavailable.

Open the complete worked design in a new tab →

The new tab keeps this draft visible. Save it on this device before closing the page.