| # Workflow mapping |
|
|
| This project reimplements the exported `JobWorkfllow` behavior without calling |
| n8n. |
|
|
| ## Input |
|
|
| The Gradio application exposes only the job description and the account-status |
| answer. Source CV, advantage, and recommendation content are loaded server-side |
| from `keys.env`. |
|
|
| ## Profile data |
|
|
| The original workflow read three n8n Data Table rows by `KEY`: |
|
|
| - `SOFTWARE_CV` |
| - `ADVANTAGE` |
| - `RECOMMENDATION_LETTER` |
|
|
| The standalone replacement loads these values from `SOURCE_CV`, `ADV`, and |
| `RECOMMENDATION_LETTER` in `keys.env`. They are not exposed as Gradio inputs. |
|
|
| ## AI branches |
|
|
| The original three agents use the connected Google Gemini Chat Model: |
|
|
| 1. Job Info Agent: Arabic structured JSON analysis. |
| 2. CV Creator Agent: factual, ATS-focused LaTeX CV, maximum two pages. |
| 3. Email Creator Agent: concise English job-application email. |
|
|
| The Python implementation runs these three Gemini calls concurrently after |
| assembling the same source inputs. The prompt constraints are kept in |
| `services/prompts.py`. |
|
|
| ## Google Sheets |
|
|
| The original Code node added the current ISO date and the Sheets node appended |
| the analysis fields. `GoogleSheetsService` preserves the same columns and |
| appends one row only when the spreadsheet configuration is complete. |
|
|
| ## PDF output |
|
|
| The generated CV remains LaTeX. The application writes the source to |
| `outputs/cv_draft.tex` and uses `pylatex` plus `pdflatex` when available to |
| produce `outputs/cv_draft.pdf`. A `reportlab` fallback keeps a downloadable |
| PDF output available when a TeX compiler is not installed. |
|
|
