InterviewsVector
Capability spine

Arc 06 · Senior AI Engineer

Model Adaptation and Multimodal Systems

Choose between prompting, retrieval, fine-tuning, preference optimization, and multimodal fusion using evidence rather than fashion.

Exit capability: Adapt a foundation model while preserving evaluation, safety, and rollback paths.

Mapped lessons
6
Published now
0
Full-arc estimate
≈15 hours
Last edited
2026-08-11

Lesson sequence

Live units open into complete labs. Planned units stay visible to show the dependency path, but intentionally have no detail route.

  1. 01

    Prompt, Retrieve, Fine-Tune, or Train?

    Planned

    Choose the smallest intervention that changes the missing capability or knowledge.

    Design reviewAdvanced95 min estimate

  2. 02

    Supervised Fine-Tuning as Behavior Shaping

    Planned

    Design data, masking, batching, and evals for instruction tuning without confusing memorization for improvement.

    Build labAdvanced125 min estimate

  3. 03

    LoRA and the Low-Rank Update Hypothesis

    Planned

    Derive parameter and memory savings, then test where low-rank adaptation stops being sufficient.

    Build labAdvanced120 min estimate

  4. 04

    Preference Optimization and Reward Failure

    Planned

    Compare preference objectives through their data assumptions and reward-hacking risks.

    Failure labAdvanced115 min estimate

  5. 05

    The Vision-Language Bridge

    Planned

    Trace images through patching, encoders, projectors, fusion, and autoregressive decoding.

    Systems labAdvanced120 min estimate

  6. 06

    Multimodal Production Boundaries

    Planned

    Engineer media ingestion, latency, privacy, safety, and modality-specific evaluation.

    Design reviewAdvanced105 min estimate