-
Notifications
You must be signed in to change notification settings - Fork 0
/
notes.txt
104 lines (88 loc) · 4.8 KB
/
notes.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
Using plantuml for diagrams.
Would be better to place diagrams in separate files and include in MD
Can preview with alt-d in real time
Need to modify settings to point to local plantuml server
We have markdown_private as a workspace so this changes paths depedning on where we open files (should not open from inside meta-markdown workspace)
Inlining UML is better for llm
Could have script that injects UML
Notebook that identifies the most important critique and proposes solutions.
The user can review inline and accept/reject
Tests are probably much easier to produce in JS rather than TS
deps.ts could allow for easy mocking with sinon
Could use a custom import_map.json and pass with --import-map
deno test --allow-all --filter "getFilePaths" tests
deno run --allow-read src/scripts/tokenCount.ts markdown_private
Build markdown features with scripts
Editorial review
Developmental Editor – Focuses on the big picture, structure, and content development.
Line Editor – Reviews sentence structure, tone, and style for flow and clarity.
Copy Editor – Corrects grammar, punctuation, and syntax on a detailed level.
Proofreader – Conducts final checks for typos and formatting errors before publishing.
Could be visual.
Content Editor – Aligns content with goals, strategy, and audience relevance.
Technical Editor – Ensures technical accuracy and clarity for specialized fields.
Fact Checker – Verifies the correctness of claims, data, and references.
Managing Editor – Oversees workflow, deadlines, and coordination in content production.
SEO Editor – Optimizes content for search engine visibility through keywords and structure.
flowchart TD
CE[Content Editor] --> DE[Developmental Editor]
DE --> TE[Technical Editor]
TE --> FC[Fact Checker]
FC --> LE[Line Editor]
LE --> SE[SEO Editor]
SE --> CE2[Copy Editor]
CE2 --> P[Proofreader]
ME[Managing Editor] -->|Coordinates throughout| CE
ME --> DE
ME --> TE
ME --> LE
ME --> CE2
Critical Review Loop Set up an automated "critique mode" that takes the current draft of a Markdown file and runs it through all these AI-powered edits sequentially. This would generate a detailed "critique report" that highlights developmental, line-level, and grammatical issues along with style suggestions, offering critical feedback without needing another human to review it.
Peer Review Simulation: AI can simulate a constructive peer review by using multiple perspectives, much like having colleagues provide feedback.
Automated Summarization and Review Reports: Using the summarization capabilities in summarizeAll.ts and getSummaryPrompt.ts, the writer can automatically generate a review summary of the entire documentation.
Meta Operation: Enable a "document structuring assistant" that helps the writer modularize their documentation, providing suggestions on breaking down long sections and organizing content more effectively.
Duplicate Text Detection and Monitoring
Define what good documentation is
System prompt
Sanity tests
End-to-end tests
UI options:
CLI https://github.com/c4spar/deno-cliffy
GUI https://github.com/webview/webview_deno
TTS
Concurrency for TTS requests to improve performance (for the podcast generation)
Disk space checks to handle potential file system issues.
Enhance text splitting for better coherence in more complex text structures.
Jupyter AI - does not work with Deno
Deno caches imports so requires restarting the kernel in notebook to refresh
Even then it seems it can cache some files sometimes ?
https://www.deeplearning.ai/short-courses/build-llm-apps-with-langchain-js/
Examples with deno + Jupyter https://github.com/rgbkrk/denotebooks
Uses Cases
----------
Maintain FAQ
Editorial review
Prioritizing feedback
Recommended edits
Complete document rewrite based on target audience
Role based presentation
Automated Glossary and Indexing
Model of the project/author
Clarifying questions
Useful
------
cloc --exclude-dir=node_modules,.git .
We don't need widgets but it is perhaps possible with
https://github.com/rgbkrk/denotebooks/blob/main/anywidget.ipynb
jupyter notebook list
jupyter kernelspec list
With VS Code we can zoom in/out (Ctrl- Ctrl+) to fit external monitor scaling on Wayland.
To run deno tasks e.g.
deno task watch-scripts
F5 runs the debugger again on the file
Lessons
-------
The notebook provides formatted output (with some limits when in VS Code)
Writing scripts may be preferrable for many tasks - with the terminal visible we have basic UI
Finding libraries with deno is painful and even more so with dependencies
We can use NPM "npm:unified" then `deno run --lock=lock.json --lock-write your_script.ts` to create a lock.json file where exact versions can be found (and I guess reused)