128 lines
3.8 KiB
Python
128 lines
3.8 KiB
Python
import marimo
|
|
|
|
__generated_with = "0.18.0"
|
|
app = marimo.App(width="medium")
|
|
|
|
|
|
@app.cell
|
|
def _():
|
|
import marimo as mo
|
|
return (mo,)
|
|
|
|
|
|
@app.cell(hide_code=True)
|
|
def _(mo):
|
|
mo.md(r"""
|
|
# Interview Analysis Pipeline Architecture
|
|
|
|
**Project Goal:** Synthesize insights from 26 stakeholder interviews into a unified report.
|
|
|
|
**Input:** 26 Interview Transcripts (`.srt`)
|
|
**Output:** Comprehensive Qualitative Analysis Report
|
|
""")
|
|
return
|
|
|
|
|
|
@app.cell(hide_code=True)
|
|
def _(mo):
|
|
mo.md(r"""
|
|
## High-Level Workflow
|
|
|
|
The analysis follows a structured **3-Stage Pipeline** to ensure consistency across all interviews while leveraging the reasoning capabilities of Large Language Models (LLMs).
|
|
""")
|
|
return
|
|
|
|
|
|
@app.cell(hide_code=True)
|
|
def _(mo):
|
|
mo.md(r"""
|
|
## Stage 1: Discovery (Theme Definition)
|
|
|
|
**Goal:** Establish the "Rules of the Game" to ensure consistent analysis.
|
|
|
|
1. **Input:** A representative sample of 4-5 interviews.
|
|
2. **Process:**
|
|
* Exploratory analysis to identify recurring topics.
|
|
* Grouping topics into **Themes**.
|
|
* Defining the **"Other"** category for emerging insights that don't fit established themes.
|
|
3. **Output:** `master_codebook.json`
|
|
* Contains Theme Names, Definitions, and Color Codes.
|
|
* Serves as the strict instruction set for the AI in Stage 2.
|
|
""")
|
|
return
|
|
|
|
|
|
@app.cell(hide_code=True)
|
|
def _(mo):
|
|
mo.mermaid("""
|
|
graph TD
|
|
A[Raw Transcripts] -->|Sample 4-5| B(Stage 1: Discovery)
|
|
B -->|Generate| C[Master Codebook]
|
|
C -->|Input| D(Stage 2: Theme Coding)
|
|
A -->|All 26 Files| D
|
|
D -->|Extract| E[Structured Dataset]
|
|
E -->|Aggregate| F(Stage 3: Synthesis)
|
|
F -->|Generate| G[Final Report]
|
|
|
|
""")
|
|
return
|
|
|
|
|
|
@app.cell(hide_code=True)
|
|
def _(mo):
|
|
mo.md(r"""
|
|
## Stage 2: Structured Theme Coding (Extraction)
|
|
|
|
**Goal:** Convert unstructured text into a structured dataset.
|
|
|
|
1. **Input:** All 26 Transcripts + `master_codebook.json`.
|
|
2. **Process:**
|
|
* The LLM analyzes each transcript segment-by-segment.
|
|
* It extracts specific quotes that match a Theme Definition.
|
|
* **Granular Sentiment Analysis:** For each quote, the model identifies:
|
|
* **Subject:** The specific topic/object being discussed (e.g., "Login Flow", "Brand Tone").
|
|
* **Sentiment:** Positive / Neutral / Negative.
|
|
3. **Output:** `coded_segments.csv`
|
|
* Columns: `Source_File`, `Speaker`, `Theme`, `Quote`, `Subject`, `Sentiment`, `Context`.
|
|
""")
|
|
return
|
|
|
|
|
|
@app.cell(hide_code=True)
|
|
def _(mo):
|
|
mo.md(r"""
|
|
## Stage 3: Synthesis & Reporting
|
|
|
|
**Goal:** Derive conclusions from the aggregated data.
|
|
|
|
1. **Input:** `coded_segments.csv` (The consolidated dataset).
|
|
2. **Process:**
|
|
* **Theme Synthesis:** All quotes for "Theme A" are analyzed together to find patterns, contradictions, and consensus.
|
|
* **"Other" Review:** The "Other" category is manually or computationally reviewed to identify missed signals.
|
|
* **Global Synthesis:** Cross-theme analysis to build the final narrative.
|
|
3. **Output:** Final Report
|
|
* Executive Summary
|
|
* Theme-by-Theme Deep Dives (with supporting quotes)
|
|
* Strategic Recommendations
|
|
""")
|
|
return
|
|
|
|
|
|
@app.cell(hide_code=True)
|
|
def _(mo):
|
|
mo.md(r"""
|
|
## Technical Infrastructure
|
|
|
|
| Component | Specification | Role |
|
|
|-----------|---------------|------|
|
|
| **Model** | `llama3.3:70b` | Primary reasoning engine (128k context) |
|
|
| **Compute** | NVIDIA H100 (80GB) | High-performance inference |
|
|
| **Orchestration** | Python + Marimo | Pipeline management and UI |
|
|
| **Storage** | Local JSON/CSV | Data persistence |
|
|
""")
|
|
return
|
|
|
|
|
|
if __name__ == "__main__":
|
|
app.run()
|