new structure for transcription engine

- found problems with mps
- integrated jinja2 templates for rendering (.tex?)
- raw ui & bad structure
This commit is contained in:
2026-02-24 02:58:28 +03:00
parent 9e67b36842
commit 3d48f473b0
24 changed files with 584 additions and 403 deletions
+58
View File
@@ -0,0 +1,58 @@
Create a complete, structured **Obsidian Markdown** educational summary based on the transcript.
The final document must be written in **{{ language }}**.
Use the transcript strictly as the factual basis:
- Do NOT invent facts.
- You may add clarifications and explanations, but they MUST be placed in special blocks.
- Preserve the logical order and meaning of the original speaker.
- Transform speech into clear academic exposition.
- Quotes must be accurate and **<= 25 words**.
- Do not reference the transcript or the speaker. Write as a stand-alone educational text.
You must follow the template below exactly.
---
# {{ title }}
## Key Ideas
- bullet points summarizing the main concepts
## Core Concepts and Definitions
- **Term** — definition
- **Another Term** — definition
## Detailed Summary
Write a logically structured, academically clear exposition.
Break it into paragraphs, subsections, lists, and quotes.
To include quotes from the transcript, use:
> "{{ quote }}"
(Ensure that every quote is ≤ 25 words.)
## Examples and Explanations
Use examples that appear in the transcript.
Additional clarifications must use the following block:
> AI clarification:
> Your extended explanation, analogy, or contextual detail.
Short notes should use:
**AI note:** short clarification
## Conclusions and Key Insights
- summary points
- what the reader should remember
## Self-Assessment Questions
- question 1
- question 2
- question 3
---
### Transcript (for your reference only):
{{ transcription }}
+75
View File
@@ -0,0 +1,75 @@
Create a clean, structured **LaTeX educational summary** based strictly on the transcript.
The final document must be written in **{{ language }}**.
Guidelines:
- Use ONLY facts from the transcript as the base.
- You MAY add explanations, but they MUST appear in designated blocks.
- Do NOT reference the transcript or the speaker.
- Maintain the speaker`s logical order.
- Produce academically clear, structured LaTeX text.
- Quotes must be accurate and <= 25 words.
Use the structure below.
Assume that stylistic environments are already defined in the `.sty` file.
---
\section*{ {{ title }} }
\subsection*{Key Ideas}
\begin{itemize}
\item main idea
\item …
\end{itemize}
\subsection*{Core Concepts and Definitions}
\begin{description}
\item[\textbf{Term}] definition
\item[\textbf{Another Term}] definition
\end{description}
\subsection*{Detailed Summary}
Write a clear, logically structured academic explanation.
Break content into paragraphs, subsections, lists.
Quotes from the transcript must be included as:
\begin{quote}
"{{ quote }}"
\end{quote}
(Ensure each quote is ≤ 25 words.)
\subsection*{Examples and Explanations}
Examples based on the transcript.
Additional clarifications must use the following custom environment (assumed to exist in the .sty):
\begin{aiclarification}
Your additional explanation, analogy, or contextual expansion.
\end{aiclarification}
Short notes should use:
\ainote{short clarification}
\subsection*{Conclusions and Key Insights}
\begin{itemize}
\item conclusion
\item conclusion
\end{itemize}
\subsection*{Self-Assessment Questions}
\begin{enumerate}
\item question
\item question
\item question
\end{enumerate}
---
% Transcript included only for reference (model must not copy or analyze or mention this directly)
% ======================================================
% {{ transcription | replace("%", "\\%") }}
+78
View File
@@ -0,0 +1,78 @@
from __future__ import annotations
import os
from jinja2 import Environment, FileSystemLoader, StrictUndefined
class PromptManager:
"""
Self-written class for Jinja2 templates especially for summarization.
Supports:
- md
- latex
Usage examples:
```
from prompts.prompt_manager import PromptManager
pm = PromptManager("notecast/prompts")
markdown_text = pm.render(
"markdown_default_prompt.md.j2",
language="Russian",
title="Calculus - Lecture 1",
transcription=raw_transcription_text
)
latex_text = pm.render(
"latex_default_prompt.j2",
language="Russian",
title="Calculus - Lecture 1",
transcription=raw_transcription_text
)
```
"""
def __init__(self, templates_dir: str) -> None:
if not os.path.isdir(templates_dir):
raise NotADirectoryError(f"Templates directory not found: {templates_dir}")
self.templates_dir = templates_dir
self.env = Environment(
loader=FileSystemLoader(self.templates_dir),
autoescape=False,
trim_blocks=True,
lstrip_blocks=True,
undefined=StrictUndefined
)
def list_templates(self) -> list[str]:
templates = []
for file in os.listdir(self.templates_dir):
if file.endswith(".j2"):
templates.append(file)
return templates
def load(self, template_name: str):
try:
return self.env.get_template(template_name)
except Exception as e:
raise RuntimeError(f"Failed to load template '{template_name}': {e}")
def render(
self,
template_name: str,
*,
language: str,
title: str,
transcription: str
) -> str:
template = self.load(template_name)
return template.render(
language=language,
title=title,
transcription=transcription
)