AI Agents I Built (6/7) — Dual-Language Blog Operation: Image Reuse and Dual Publishing

A pipeline design that reduces cost and redundant work when publishing the same content to Korean and English blogs simultaneously.

What This Post Covers

  • How to configure a Blogger API v3 dual-publishing pipeline with a single --blog flag for Blog ID switching
  • The conditions and automation script structure that bring English blog image generation cost to $0 via image reuse
  • A ratio-based mapping algorithm for placing images correctly when Korean and English posts differ in section count
  • The Blogger API post size ceiling when using base64 inline images, and mitigation options

Background: Why Dual Operation Needs Automation

Korean technical blogs have a hard audience ceiling set by population. Topics like AI agents, Claude Code, and local LLMs have disproportionately larger search demand in English, making bilingual publishing an effective audience expansion strategy.

The problem: maintaining both languages doubles every step — writing, image generation, publishing, and table of contents management. This post documents an implementation across four axes (pipeline branching, rewrite strategy, image reuse, position mapping) that eliminates most of that duplication cost.

1. Dual Publishing Pipeline — One Flag, One Code Path

Adding a --blog flag to blogger_publish.py allows publishing to both blogs through the same OAuth token and the same code path, switching only the Blog ID.

  • Korean: Blog ID 8353076619669124413 (maju-not.blogspot.com)
  • English: Blog ID 2499508976191968092 (maju-tech.blogspot.com)

Blogger API v3 does not segregate auth scopes or endpoints by blog. A single token grants access to all owned blogs, so authentication does not need to be duplicated. Publish, update, delete, and list operations all run through a single code path.

Publishing flow: 1. Write Korean post → save to drafts/blog/ 2. Publish Korean: python3 blogger_publish.py draft.md --blog kr 3. Save English rewrite 4. Publish English: python3 blogger_publish.py draft_en.md --blog en

2. English Posts — Rewriting, Not Translating

Korean and English technical writing follow different stylistic conventions. Korean technical prose conventionally uses formal polite endings; English technical writing favors active voice, short sentences, and direct instruction. Machine translation flattens this difference and produces unnatural output.

The unit of work is therefore "preserve structure and argument + rewrite in style", not translation. Section structure and claims from the original are retained, but prose is rebuilt sentence-by-sentence to match English technical blog conventions. During this process:

  • Korean-specific service names and community references are replaced with globally intelligible equivalents
  • Indirect phrasing natural in Korean becomes direct assertion
  • Background context meaningful only to Korean readers (domestic policy, local market specifics) is condensed or removed in the English version

3. Image Reuse — Conditions for $0 Additional Cost

This is the largest savings. Generating 228 images for 76 Korean blog posts with Nano Banana 2 cost approximately $15. Reusing the same images for the English blog brings image generation cost to $0.

Three preconditions must hold for image reuse to work:

  • All prompts were written in English from the start: Korean-language prompts can produce Korean text in generated images
  • All text within images is in English: diagram labels, speech bubbles, and captions must all be in English
  • Content is predominantly technical diagrams: abstract and technical visuals that do not depend on cultural context are language-neutral and reusable

If any precondition breaks — Korean captions, culturally specific illustrations — image reuse is not viable and separate English assets must be generated.

The copy_images_to_en.py automation script: 1. Fetch post lists from both the Korean and English blogs 2. Match Korean and English posts by publication order 3. Extract base64 image divs from Korean post HTML 4. Insert identical images at corresponding positions in English post HTML 5. Update via Blogger API

Processing all 36 English posts takes approximately 2 minutes. The bottleneck is API call rate, not image generation — because image generation is absent from this step entirely.

4. Ratio-Based Position Mapping — When Section Structure Differs

During rewriting, the English post's section count or order may diverge from the original. Mechanically inserting images at "the same index" produces misaligned context.

The solution is heading-based ratio mapping:

  • Calculate the ratio position of each image relative to total h2/h3 headings in the Korean post
  • Multiply that ratio by the English post's heading count to determine insertion position

Example: if a Korean post has 10 sections and an image appears before the 3rd section, ratio = 30%. If the English post has 8 sections, the image is placed before the 8 × 0.30 ≈ 2nd section. The alignment is not exact, but intro-region images land near the intro and conclusion-region images land near the conclusion, preserving contextual coherence.

5. Cost Structure

Item Korean Blog English Blog Total
Image generation $15.3 $0 $15.3
Blogger API Free Free Free
Domain blogspot (free) blogspot (free) Free
Content writing Claude API Claude API Claude API cost only

When image reuse preconditions hold, the marginal cost of operating an additional English blog converges to Claude API call cost for the rewrite. Additional cost for images, hosting, and publishing infrastructure is $0.

Limitations and Improvement Directions

Actual rewrite cost: Machine translation cannot reliably meet quality standards, so each English post requires effort roughly equivalent to rewriting a substantial portion of the original. A two-stage pipeline — translation API for a first draft, then style correction with an LLM — is a potential alternative.

Post-order matching fragility: The current image copy script matches Korean and English posts by publication order. Adding posts to only one blog misaligns the sequence and causes images to be inserted into the wrong posts. The improvement path is either title-similarity matching (embedding-based) or an explicit kr_post_id → en_post_id mapping table maintained at publish time.

Base64 inline image size ceiling: Inlining images as base64 inflates post HTML rapidly. Distributing 228 images across posts resulted in some post HTML exceeding 2 MB. Blogger API enforces a per-request post size limit of approximately 1 MB, causing update failures on affected posts. Mitigation options: (1) upload images to an external CDN and replace inline data with URL references; (2) cap images per post; (3) reduce base64 payload size via image resize and WebP conversion.

Scope and Open Questions

This structure extends directly to the general problem of scaling the same content across N languages. Expanding the --blog flag from kr/en to kr/en/ja and maintaining image reuse preconditions (English prompts, language-neutral diagrams) brings Japanese blog image cost to $0 as well.

Two open questions remain: - How far can the rewrite process be automated — does a two-stage pipeline (translation API + style-correction LLM) approach the quality of manual rewriting? - When switching post matching to title similarity, how should the similarity threshold be calibrated to correctly exclude posts that exist on only one blog?

Resolving both would bring bilingual operation to the level of "write once, publish twice."

Series overview: Series index

๋Œ“๊ธ€

๋Œ“๊ธ€ ์“ฐ๊ธฐ

์ด ๋ธ”๋กœ๊ทธ์˜ ์ธ๊ธฐ ๊ฒŒ์‹œ๋ฌผ

"ML Foundations (9/9) — PyTorch vs TensorFlow, and the Road to Local LLMs"

์ด๋ฏธ์ง€
← 8/9 Deep Learning Architec… ๐Ÿ“š Series Index (series end) The final part. So far we've focused on models. Now we focus on the tools that actually run them . We'll lay out the philosophical difference between PyTorch and TensorFlow, and trace how Hugging Face transformers , llama.cpp , MLX , and Ollama built the bridge to running large language models on your own machine. By the end you should have the full mental model of "download a pretrained LLM and serve it locally." 0. Learning Objectives Explain the eager-vs-graph difference between PyTorch and TensorFlow. Explain, in graph terms, how autograd automates the backward pass. Use Hugging Face transformers ' AutoModel , AutoTokenizer , and pipeline abstractions. Describe llama.cpp's GGUF quantization, INT4 inference, and CPU-first flow. Describe Apple MLX's use of unified memory and how it differs from PyTorch. Run a local LLM with Ollama and call it through an OpenAI-compatible API. ...
Read more »

Agent Memory Engine (2/10) — Building an AI Agent Memory System with SQLite Alone

← 1/10 SQLite Memory Engine f… ๐Ÿ“š Series Index 3/10 memcore Completed → A memory engine that runs without daemons, without dependencies, anywhere ํ•ต์‹ฌ ์š”์•ฝ This post covers how to design an AI agent memory engine (memcore) on a single-file SQLite backend. - Structure: 25 modules, ~6,464 lines, 22 DB tables - Principles: no daemon, single-file portability, no required external dependencies, host-agnostic - Core techniques: 4-layer hybrid search (topic · vector · FTS5 · LIKE), U-tag dialectic-based confidence evolution, ingestion-layer bias rejection rules What This Post Covers One approach to building AI agent memory without a server or cloud vector DB. Specifically: (1) the limitations of text-file-based memory, (2) vector DB alternatives compared and selection criteria, (3) how to layer search tiers, (4) how to evolve confidence scores over time, and (5) how to block biased memory accumulation at the ingestion stage. Design Principles: Why Single-File SQLite Text-file-based ...
Read more »

"ML Foundations (8/9) — Deep Learning Architectures: CNN, RNN, Attention"

์ด๋ฏธ์ง€
← 7/9 Deep Learning Training ๐Ÿ“š Series Index 9/9 PyTorch vs TensorFlow → What did we keep adding on top of an MLP? Vision went CNN, sequences went RNN/LSTM, and eventually both converged on Attention and the Transformer. Each architecture is easier to remember if you read it as what it gave up to gain something else . This part is the one-line summary of the decisive inflection points: LeNet → AlexNet → ResNet → LSTM → Attention → Transformer. 0. Learning Objectives Explain why CNN's convolution, pooling, and weight sharing match images so well. Trace what got unblocked from LeNet → AlexNet → VGG → ResNet as depth grew. State BPTT for RNNs and the vanishing/exploding gradient problem. Write the LSTM gate equations (forget, input, output) with intuition. Write the attention formula in query/key/value form and the scaled dot-product variant. State the precise reasons Transformers replaced RNNs (parallelism + long-range dependencies). 1. ํ•ต์‹ฌ ์š”์•ฝ CNN : weight sharin...
Read more »

"RAG Core Study (14/26) — Evaluation Sets with RAGAS & DeepEval"

์ด๋ฏธ์ง€
Series overview: Series index ๐Ÿ“š Series Index Without an evaluation set, every RAG improvement is a story, not evidence. RAG systems fail in more than one place: retrieval, context selection, answer generation, and grounding. That means "looks better" is not a metric. Part 14 explains how to build a golden dataset , how RAGAS and DeepEval fit into the workflow, and why teams must separate retrieval quality from answer quality if they want tuning decisions to hold up. 0. Prerequisites Part 13 reranking — system quality now depends on multi-stage ranking. Part 1 grounding — a correct-looking answer is not enough. Part 12 Hybrid — multiple retrieval settings need comparison on fixed data. 1. Learning Objectives Build a small but useful golden dataset for RAG. Distinguish retrieval metrics from answer metrics . Use RAGAS and DeepEval in the right roles. Avoid the common traps in judge-model-based evaluation. 2. ํ•ต์‹ฌ ์š”์•ฝ An evaluation set for RAG is ...
Read more »

"ML Foundations (7/9) — Deep Learning Training: Optimizers, Regularization, Initialization"

← 6/9 Neural Networks ๐Ÿ“š Series Index 8/9 Deep Learning Architec… → Building an MLP, as we did in Part 6, is not the same as getting it to train well . The right optimizer, regularization, initialization, and learning rate — when these line up, deep networks converge. When they don't, the network refuses to learn at all. This part is the catalog of those crafts, with formulas, paper citations, and code in one place. 0. Learning Objectives Compare and write the update rules for SGD, Momentum, Nesterov, and Adam. Explain how Dropout, BatchNorm, and LayerNorm work and where they belong in a model. Derive the variance formulas for Xavier and He initialization and match them to activations. Implement step, cosine, and warmup learning-rate schedules in PyTorch. Explain why gradient clipping is effectively required for RNNs and Transformers. Diagnose the most common training failures (NaN, plateau, overfitting) and apply first-line fixes. 1. ํ•ต์‹ฌ ์š”์•ฝ SGD : \(w \leftarro...
Read more »

AI Agents I Built (5/7) — Building an Automated Blogger API Publishing System

← 4/7 My Life Organizer Agent ๐Ÿ“š Series Index 6/7 Dual → Designing a Markdown-Based Auto-Publish Pipeline with Blogger API v3 + OAuth 2.0 Key Summary Blogger API v3 + OAuth 2.0 enables a pipeline that converts Markdown files into publishable HTML posts blogger_publish.py handles the full flow — frontmatter parsing → body cleanup → HTML conversion → CSS injection → API post — in a single script For bulk publishing, quota management (delay, retry, TOC suppression) is the critical design concern Platform Selection Criteria API access is mandatory for AI agent-driven content publishing. A comparison of platforms: Platform API Cost AdSense Decision WordPress.com REST API Paid plan required Paid only Cost overhead Ghost REST API Self-hosted or paid Manual setup Ops overhead Blogger v3 API Free Native integration Adopted Blogger is free, exposes a REST API, and integrates natively with AdSense. For personal blog automation, this combination i...
Read more »

"ML Foundations (6/9) — Neural Networks: From Perceptron to MLP"

์ด๋ฏธ์ง€
← 5/9 Regularization & Model… ๐Ÿ“š Series Index 7/9 Deep Learning Training → A neural network is logistic regression from Part 4 stacked into layers. Each layer passes through a new nonlinear activation, expanding the family of functions the network can represent. In this part we put how a neural network learns onto one page. Perceptrons, XOR, MLPs, backpropagation, activation functions, and the universal approximation theorem all meet here. 0. Learning Objectives Write the perceptron's learning rule and explain why it cannot solve XOR. Express an MLP's forward pass as one line of matrix products. Derive backpropagation by hand on a two-layer MLP. Compare the definitions, derivatives, and trade-offs of sigmoid, tanh, ReLU, and GELU. State the universal approximation theorem (Cybenko 1989 / Hornik 1991) precisely. Train an MLP end-to-end in PyTorch. 1. ํ•ต์‹ฌ ์š”์•ฝ Perceptron (1958): \(\hat{y} = \mathrm{step}(w^\top x + b)\). Converges only on linearly separabl...
Read more »