Post 4 of the Pepper & Carrot AI flipbook series. The comic is images, not text — so before any RAG can happen, every page needs a description. This post walks through using a Claude Code skill as the vision provider for the ingestion pipeline of this portfolio-project specifically: no per-call API cost beyond the Claude Code subscription, auditable JSON artifacts on disk, same Claude model as Anthropic's hosted vision API. By the end, one full episode is ingested into Postgres + ChromaDB + local storage. The right vision provider is context-specific — local VLM, hosted API, and Claude Code each win under different constraints (budget, whether the pipeline runs unattended, throughput) — and the post includes a decision matrix mapping each constraint to the right choice.

Post 3 of the Pepper & Carrot AI flipbook series. Build three typed Protocol interfaces — Storage, EmbeddingClient, ChatClient — and the factory that picks the right implementation from a .env file. By the end you have LocalStorage serving images end-to-end and a working embedding client producing real 1024-dim vectors against the local Ollama you set up in Post 2.

Post 2 of the Pepper & Carrot AI flipbook series. Stand up the local workshop: Postgres in Docker, Ollama serving qwen2.5:7b and bge-m3, a typed FastAPI scaffold that passes mypy strict, the first Alembic migration, and one downloaded episode on disk. By the end you have the workshop every later post is built on.

Series introduction for the Pepper & Carrot AI-powered flipbook — a page-aware, spoiler-safe RAG app for a hand-drawn webcomic that runs locally on a laptop or on a five-piece free-tier cloud deploy. What it is, why it's a different problem from chat-with-PDF, and what each post in the series builds.

Applying Anthropic's Contextual Retrieval technique to personal photo libraries — where the notion of a "chunk" is a photo, not a paragraph.

An engineer's hands-on experiment connecting Claude to Xiaohongshu via MCP — and what it reveals about the future of AI-powered content creation.

A hands-on guide to training a math reasoning model with GRPO on Lambda Cloud using 2xH100 GPUs — improving Qwen2.5-Math-1.5B accuracy from ~6% to ~25% with practical implementation details.

A beginner's guide to reinforcement learning for language models — breaking down policy gradients, baselines, importance sampling, PPO, and GRPO in plain English with step-by-step examples.

Fine-tuning Qwen3-1.7B on the MATH dataset using supervised fine-tuning (SFT) on Lambda Labs, bridging the gap from local MacBook development to cloud GPU training.

A practical guide to developing ML training code on a MacBook with Apple Silicon, then seamlessly deploying to Google Colab or multi-GPU cloud instances like Lambda Labs.