algernon-synthesis
You run a cross-material synthesis session. The goal is to build explicit
connections between concepts learned in different materials — the kind of
holistic understanding that separates someone who memorized facts from
someone who can actually design systems.
Constants
CODEBLOCK0
Step 1 — Check Eligibility
CODEBLOCK1
If fewer than 2 materials have reviews:
"Synthesis requires at least 2 studied materials. Study more material first."
Step 2 — Identify Cross-Material Concept Overlaps
From the tags and topics of reviewed cards across all studied materials,
identify 3-5 concept pairs that appear in multiple materials but may be
understood differently in each context.
Examples of strong synthesis pairs:
- - "evaluation" in RAG vs LLMOps contexts
- "chunking" in embedding vs RAG contexts
- "latency" in inference vs retrieval contexts
- "context" in prompt engineering vs agent memory contexts
- "retrieval" in BM25 vs vector similarity vs caching contexts
Prefer pairs where the same word genuinely means something different in
each context — that contrast is the richest learning opportunity.
Step 3 — Synthesis Questions
For each concept pair, ask:
AskUserQuestion (free text):
"[CONCEPT] appears in both [MATERIALA] and [MATERIALB]. How does the meaning
or role of [CONCEPT] differ between these two contexts? Where do they overlap?"
After each answer, give brief feedback:
- - Name what the user connected well.
- Name any distinction they missed (without lecturing — one sentence).
Step 4 — Production Scenario Challenge
AskUserQuestion (free text):
"If you were building a production AI system, how would the knowledge from
[MATERIALA] and [MATERIALB] work together? Give a concrete scenario with
specific design decisions."
Evaluate for:
- 1. Coherence — does the scenario make technical sense?
- Specificity — are there real design decisions, not just buzzwords?
- Correct use of concepts — are terms from both materials used accurately?
Step 5 — Summary
Display:
CODEBLOCK2
Send to Notion
Send to the Notion page of the most recent phase studied:
CODEBLOCK3
Include:
- - Cross-material concepts explored
- Gaps identified (bridges that need reinforcement)
- The production scenario the user described
Save Memory
Append to today's conversation log:
CODEBLOCK4
algernon-synthesis
你正在进行一次跨材料综合学习。目标是建立不同材料中所学概念之间的显式连接——这种整体性理解，正是区分死记硬背者与真正能设计系统者的关键。
常量
DB=/home/antonio/Documents/huyawo/estudos/vestibular/data/vestibular.db
NOTION_CLI=~/go/bin/notion-cli
第一步——检查资格
bash
sqlite3 $DB \
SELECT m.slug, m.name, COUNT(r.id) as review_count
FROM materials m
JOIN decks d ON d.material_id = m.id
JOIN cards c ON c.deck_id = d.id
JOIN reviews r ON r.card_id = c.id
GROUP BY m.id
HAVING review_count > 0
ORDER BY review_count DESC;
如果少于2份材料有复习记录：
综合学习至少需要2份已学习材料。请先学习更多材料。
第二步——识别跨材料概念重叠
从所有已学习材料的复习卡片标签和主题中，识别出3-5对出现在多份材料中但在不同上下文中可能被不同理解的概念对。
强综合对的示例：
- - RAG与LLMOps上下文中的评估
- 嵌入与RAG上下文中的分块
- 推理与检索上下文中的延迟
- 提示工程与智能体记忆上下文中的上下文
- BM25与向量相似度及缓存上下文中的检索
优先选择那些同一词汇在不同上下文中确实含义不同的概念对——这种对比是最丰富的学习机会。
第三步——综合问题
对于每个概念对，提问：
AskUserQuestion（自由文本）：
[概念]同时出现在[材料A]和[材料B]中。在这两种上下文中，[概念]的含义或作用有何不同？它们在哪里有重叠？
每次回答后，给出简短反馈：
- - 指出用户连接得好的地方。
- 指出他们遗漏的任何区别（不加说教——一句话即可）。
第四步——生产场景挑战
AskUserQuestion（自由文本）：
如果你正在构建一个生产级AI系统，[材料A]和[材料B]的知识将如何协同工作？请给出一个具体场景，包含具体的设计决策。
评估标准：
- 1. 连贯性——场景在技术上是否合理？
- 具体性——是否有真实的设计决策，而不仅仅是流行词汇？
- 概念的正确使用——两份材料中的术语是否被准确使用？
第五步——总结
显示：
综合学习会话完成。
涵盖材料：[列表]
良好建立的概念桥梁：[列表]
需要加强的桥梁：[列表]
发送到Notion
发送到最近学习阶段的Notion页面：
bash
~/go/bin/notion-cli append --page-id PHASEPAGEID --content MARKDOWN
包含：
- - 探索的跨材料概念
- 发现的差距（需要加强的桥梁）
- 用户描述的生产场景
保存记忆
追加到今日对话日志：
[HH:MM] 综合学习会话
材料：[列表] | 已建桥梁数：N | 需要加强：[列表]
