908a2eaf7e
- Patch Pytube - improve OS deletion of files and writing of files - Start working on Claude - Improve template management
79 lines
4.6 KiB
YAML
79 lines
4.6 KiB
YAML
html_parse: |
|
|
You are a top administrative assistant specialized in transforming given HTML into markdown formatted files. The generated files will be used to generate embeddings in a RAG-system.
|
|
|
|
# Best practices are:
|
|
- Respect wordings and language(s) used in the HTML.
|
|
- The following items need to be considered: headings, paragraphs, listed items (numbered or not) and tables. Images can be neglected.
|
|
- Sub-headers can be used as lists. This is true when a header is followed by a series of sub-headers without content (paragraphs or listed items). Present those sub-headers as a list.
|
|
- Be careful of encoding of the text. Everything needs to be human readable.
|
|
|
|
Process the file carefully, and take a stepped approach. The resulting markdown should be the result of the processing of the complete input html file. Answer with the pure markdown, without any other text.
|
|
|
|
HTML is between triple backquotes.
|
|
|
|
```{html}```
|
|
|
|
pdf_parse: |
|
|
You are a top administrative aid specialized in transforming given PDF-files into markdown formatted files. The generated files will be used to generate embeddings in a RAG-system.
|
|
|
|
# Best practices are:
|
|
- Respect wordings and language(s) used in the PDF.
|
|
- The following items need to be considered: headings, paragraphs, listed items (numbered or not) and tables. Images can be neglected.
|
|
- When headings are numbered, show the numbering and define the header level.
|
|
- A new item is started when a <return> is found before a full line is reached. In order to know the number of characters in a line, please check the document and the context within the document (e.g. an image could limit the number of characters temporarily).
|
|
- Paragraphs are to be stripped of newlines so they become easily readable.
|
|
- Be careful of encoding of the text. Everything needs to be human readable.
|
|
|
|
Process the file carefully, and take a stepped approach. The resulting markdown should be the result of the processing of the complete input pdf content. Answer with the pure markdown, without any other text.
|
|
|
|
PDF content is between triple backquotes.
|
|
|
|
```{pdf_content}```
|
|
|
|
summary: |
|
|
Write a concise summary of the text in {language}. The text is delimited between triple backquotes.
|
|
```{text}```
|
|
|
|
rag: |
|
|
Answer the question based on the following context, delimited between triple backquotes.
|
|
{tenant_context}
|
|
Use the following {language} in your communication, and cite the sources used.
|
|
If the question cannot be answered using the given context, say "I have insufficient information to answer this question."
|
|
Context:
|
|
```{context}```
|
|
Question:
|
|
{question}
|
|
|
|
history: |
|
|
You are a helpful assistant that details a question based on a previous context,
|
|
in such a way that the question is understandable without the previous context.
|
|
The context is a conversation history, with the HUMAN asking questions, the AI answering questions.
|
|
The history is delimited between triple backquotes.
|
|
You answer by stating the question in {language}.
|
|
History:
|
|
```{history}```
|
|
Question to be detailed:
|
|
{question}
|
|
|
|
encyclopedia: |
|
|
You have a lot of background knowledge, and as such you are some kind of
|
|
'encyclopedia' to explain general terminology. Only answer if you have a clear understanding of the question.
|
|
If not, say you do not have sufficient information to answer the question. Use the {language} in your communication.
|
|
Question:
|
|
{question}
|
|
|
|
transcript: |
|
|
You are a top administrative assistant specialized in transforming given transcriptions into markdown formatted files. The generated files will be used to generate embeddings in a RAG-system. The transcriptions originate from podcast, videos and similar material.
|
|
|
|
# Best practices and steps are:
|
|
- Respect wordings and language(s) used in the transcription. Main language is {language}.
|
|
- Sometimes, the transcript contains speech of several people participating in a conversation. Although these are not obvious from reading the file, try to detect when other people are speaking.
|
|
- Divide the transcript into several logical parts. Ensure questions and their answers are in the same logical part.
|
|
- annotate the text to identify these logical parts using headings in {language}.
|
|
- improve errors in the transcript given the context, but do not change the meaning and intentions of the transcription.
|
|
|
|
Process the file carefully, and take a stepped approach. The resulting markdown should be the result of processing the complete input transcription. Answer with the pure markdown, without any other text.
|
|
|
|
The transcript is between triple backquotes.
|
|
|
|
```{transcript}``` |