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
- 6
- Full-arc estimate
- ≈15 hours
- Last edited
- 2026-09-19
Lesson sequence
Live units open into complete labs. Planned units stay visible to show the dependency path, but intentionally have no detail route. A withdrawn unit is retained only at its previously advertised URL and is not presented as live.
- 01
Prompt, Retrieve, Fine-Tune, or Train?
LiveChoose the smallest intervention that changes the missing capability or knowledge.
Design reviewAdvanced95 min estimateArtifact: Adaptation decision record
- 02
Supervised Fine-Tuning as Behavior Shaping
LiveDesign data, masking, batching, and evals for instruction tuning without confusing memorization for improvement.
Build labAdvanced125 min estimateArtifact: Supervised fine-tuning data contract
- 03
LoRA and the Low-Rank Update Hypothesis
LiveDerive parameter and memory savings, then test where low-rank adaptation stops being sufficient.
Build labAdvanced120 min estimateArtifact: Low-rank adaptation budget
- 04
Preference Optimization and Reward Failure
LiveCompare preference objectives through their data assumptions and reward-hacking risks.
Failure labAdvanced115 min estimateArtifact: Preference evidence audit
- 05
The Vision-Language Bridge
LiveTrace images through patching, encoders, projectors, fusion, and autoregressive decoding.
Systems labAdvanced120 min estimateArtifact: Vision-language bridge contract
- 06
Multimodal Production Boundaries
LiveEngineer media ingestion, latency, privacy, safety, and modality-specific evaluation.
Design reviewAdvanced105 min estimateArtifact: Multimodal production release gate