A governed resume-tailoring system designed to refuse low-confidence output.
[License: Dual MPL-2.0/Commercial]
Threshold is a governed resume-tailoring system designed around AI failure modes. In the app, tailoring only runs when job compatibility meets a configurable threshold, and generated content is checked against the source resume for potential grounding issues.
Most resume tools say "generate better resumes." Threshold says the system can refuse to generate. Three rules govern that:
- Threshold gate. In the app, tailoring runs only when the match score meets the configured threshold (default 80%). Below it, the Tailor button stays locked, so a weak fit cannot be tailored by accident. The gate is currently enforced in the app interface; backend enforcement is planned work. A resume tool that tells you no.
- Factual grounding. The AI is instructed to use only information from the source resume, and the prompt includes a self-check asking it to remove anything not supported by the original before outputting. That is an instruction to the model, not a guarantee about the output.
- Output constraints. Tailored resumes follow a fixed structure (standard sections, single column, no tables) and are designed to fit 2 pages.
The AI proposes. Rules and humans dispose.
Grouped by the rule each one supports:
- Configurable minimum match threshold (slider 50-95%, default 80%)
- Compatibility scoring returns a score, a verdict (STRONG, MODERATE, or WEAK), evidence, and gaps
- Tailoring stays disabled until the score meets the threshold, and editing the resume, job description, or threshold invalidates a stored passing score
- When the gate passes, the app tailors the resume and cover letter
- The AI is instructed to use only information from the source resume
- A self-check in the prompt asks the model to remove anything not supported by the original before outputting
- The prompt forbids invented experience and overselling
- Fixed structure: standard sections, single column, no tables
- Designed to fit 2 pages; if a PDF runs over, the app offers to condense it with grounded cuts you approve
- Export as PDF, Word, plain text, or ATS-optimized
- You review the score, verdict, evidence, and gaps before tailoring runs
- You review the tailored resume and cover letter before you use them
- When a PDF runs over 2 pages, you choose which cuts to approve
- Job import from LinkedIn, email, or plain text, with job descriptions stored for later
- Role-based customization: Standard, Senior, Lead, Principal
- Custom prompt templates with validation, built-in examples, and variable preview
- Modern themed interface
First evaluation published 2026-11-05 (docs/evaluation_results.md): 1 sample, a condensed resume summary against a sample Senior Manager, Technical Operations job description. Overall score 95, verdict STRONG, 3 evidence items, 0 gaps, and the threshold gate passed (95 >= 80). Grounding held: nothing in the output came from outside the source resume.
One sample is a start, not a benchmark. The next evaluations should cover a borderline posting and a clear mismatch, to test the WEAK path and below-threshold refusal.
pip install -r requirements.txtcp .env.example .env
# Edit .env and add your GEMINI_API_KEY from https://aistudio.google.com/apikeypython gui/tkinter_app.pyThe Threshold application icon.
flowchart TD
A[Resume + Job Description] --> B[Compatibility Analysis]
B --> C{Score meets threshold?}
C -- No --> D[Refuse: no tailoring]
C -- Yes --> E[Tailoring Engine]
E --> F[Grounding self-check]
F --> G[Output: tailored resume and cover letter]
- Paste the job description (minimum 100 characters)
- Click "Analyze Match" to get your compatibility score
- Review detailed feedback on strengths and improvement areas
- For scores meeting or exceeding the configured threshold (default 80%), the "Start Tailoring" button becomes available
- AI generates a customized resume and cover letter
- All materials are automatically saved to the output folder
- Use the "Import Job" tab to import job descriptions from:
- Plain text
- LinkedIn job postings (HTML source)
- Email content
- Automatically parses job titles, companies, and descriptions
- Use the "Custom Prompts" tab to create and manage custom prompt templates
- Load built-in examples as starting points for your custom prompts
- Preview how variables will be substituted in your prompts
- Save custom templates for specific roles or companies
- View previously tailored applications in the "Tailored Documents" tab
- Export your materials in multiple formats:
- PDF (professional formatting)
- Word (.docx)
- Plain Text (.txt)
- ATS-Optimized (for Applicant Tracking Systems)
- Resumes are designed to fit 2 pages. If a PDF export runs over, the app offers to condense it: the AI suggests specific cuts or rewrites (oldest and least relevant first), you check the ones you want, and the revised resume is saved back automatically.
- Delete tailored documents selectively:
- Delete only resume and cover letter (job description preserved by default)
- Option to delete all files including job description
Adjust the minimum compatibility score required to enable tailoring:
Access the Settings tab in the application to configure the match threshold:
- Slider Control: Drag the slider to adjust the threshold from 50% to 95%
- Manual Entry: Enter a specific percentage value in the text field
- Real-time Feedback: Current threshold value is displayed dynamically
- Apply Changes: Click "Apply Changes" to save your custom threshold
- Reset to Default: Use "Reset to Default (80%)" to restore the default setting
Edit the value in config/settings.py:
# In config/settings.py
MIN_MATCH_THRESHOLD = 80 # Default: 80%The match threshold determines when the "Start Tailoring" button becomes available:
- Match Score ≥ Threshold: Tailoring is enabled
- Match Score < Threshold: Tailoring is disabled to prevent poor-quality results
80 is the starting default. It has not been validated: the scoring prompt gives the model no rubric and no score bands, and the only published evaluation scored 95, so the gate has never been tested near 80. Treat 80 as a starting point, not a measured cutoff. Calibration against labeled postings is planned work; see docs/score-calibration.md. Until it is done, no threshold value carries a recommendation.
- Start with the default 80% threshold
- Adjust based on your job application results
- Monitor the detailed match analysis to understand score factors
Choose the appropriate role level for optimal tailoring:
- Standard: Entry to mid-level positions (0-5 years)
- Senior: Senior-level positions (5-10 years)
- Lead: Lead positions (8-15 years)
- Principal: Principal/architect positions (12+ years)
Select the level that best matches the position requirements.
# Install PyInstaller
pip install pyinstaller
# Build cross-platform executable
python build_standalone.pyCommon issues and solutions:
- "API key not found": Ensure .env file contains your GEMINI_API_KEY
- "No active resume": Upload and set a resume as active in the Resume Management tab
- "Match analysis failed": Job description must be at least 100 characters
Threshold is available under two licensing options:
- Free to use, modify, and distribute
- Modifications must be open-sourced
- Perfect for individual users and open-source projects
- Paid license required for proprietary use
- No requirement to open-source modifications
- Includes priority support and additional features
- Ideal for businesses and enterprise use
For commercial licensing inquiries, please contact via https://github.com/Micouwr.
See LICENSE_MPL.md and LICENSE_COMMERCIAL.md for complete license terms.
