Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/tech-dipesh/yeti-Jobs/llms.txt

Use this file to discover all available pages before exploring further.

When you upload a resume to Yeti Jobs, it doesn’t just get stored — it gets read. The platform extracts the full text from your PDF using pdf-parse, sends it to a Groq-hosted AI model with a structured prompt, and returns a numeric ATS (Applicant Tracking System) compatibility score alongside a set of concrete improvement suggestions. The score and feedback are persisted in the ats_scores table and linked back to your user account so you can retrieve them at any time without re-uploading.

​
How the Scoring Pipeline Works

1

Upload your resume PDF

Send a multipart/form-data POST request to the resume endpoint with your PDF attached under the resume field. The server validates the file is present before proceeding.
curl -X POST https://yeti-jobs.onrender.com/api/v1/users/resume \
  -H 'Cookie: token=<your_jwt_token>' \
  -F 'resume=@/path/to/your-resume.pdf'
2

Text extraction via pdf-parse

The backend instantiates a PDFParse parser with the raw file buffer and calls getText(). This extracts all readable text from your PDF without requiring a file to be written to disk, keeping the operation fully in-memory.
3

AI analysis via Groq API

The extracted text is passed to the gemini utility (backend/src/utils/grok.js), which calls the Groq API using an OpenAI-compatible client. The call uses client.responses.create with model "openai/gpt-oss-20b" against base URL https://api.groq.com/openai/v1. The prompt instructs the model to return only a raw JSON object — no markdown, no code fences, no explanatory prose — with exactly two keys: ats_scores and feedback.The environment variable GROK_API must be set to a valid Groq API key for this step to succeed.
4

Parallel storage operations

Using Promise.all, the backend simultaneously:
  • Uploads the PDF to Supabase Storage (resume bucket) and gets the public URL.
  • Inserts a new row into ats_scores with the score (integer from ats_scores key) and feedback (JSONB array).
  • Updates the user’s resume_url column in the users table.
If any prior resume exists, the old Supabase object is deleted and the previous ats_scores row is removed before the new data is written. All writes are wrapped in a database transaction (BEGIN / COMMIT / ROLLBACK).

​
Upload Endpoint

POST /api/v1/users/resume
The request must be multipart/form-data. The file field name must be resume. 
curl -X POST https://yeti-jobs.onrender.com/api/v1/users/resume \
  -H 'Cookie: token=<your_jwt_token>' \
  -F 'resume=@/Users/alex/Documents/alex-resume-2025.pdf'
On success the API responds with 201 Created: 
{
  "message": "Resume Uploaded Successfully"
}

​
Retrieving Your Score

Fetch your most recent resume URL and ATS analysis result. The query joins ats_scores with users and returns only the latest record ordered by created_at DESC. 
GET /api/v1/users/resume

curl https://yeti-jobs.onrender.com/api/v1/users/resume \
  -H 'Cookie: token=<your_jwt_token>'

​
Response Format

The API returns the data stored in the ats_scores table, joined with the resume_url from users: 
{
  "message": {
    "user_id": "b7e1a2f3-0000-4cde-9876-000000000099",
    "created_at": "2025-07-10T14:22:05.000Z",
    "score": 72,
    "feedback": [
      "Add quantifiable achievements to work experience",
      "Include relevant keywords matching the job description",
      "Improve the skills section formatting"
    ],
    "resume_url": "https://<supabase-project>.supabase.co/storage/v1/object/public/resume/upload/<uuid>-resume.pdf"
  }
}
FieldTypeDescription
scoreinteger (0–100)ATS compatibility score. Higher is better.
feedbackstring arrayOrdered list of actionable improvement points.
resume_urlstringPublic URL to your uploaded PDF in Supabase Storage.
created_attimestampWhen this analysis was performed.

​
Database Schema

Scores are stored in the ats_scores table: 
CREATE TABLE ats_scores (
  uid         UUID PRIMARY KEY DEFAULT gen_random_uuid(),
  user_id     UUID REFERENCES users(uid) ON DELETE CASCADE,
  created_at  TIMESTAMPTZ DEFAULT now(),
  score       INT4,
  feedback    JSONB
);
The feedback column is stored as JSONB, which allows efficient querying and indexing of individual feedback points if needed in the future.

​
AI Prompt Structure

The Groq API is called via client.responses.create with model "openai/gpt-oss-20b". The prompt enforces strict JSON-only output with no markdown or prose: 
Must Not Use Any Ai Words
Analyze this resume and return ONLY a valid JSON object with no markdown,
no backticks, no explanation.
Strictly follow this format:
{
  "ats_scores": <number between 0-100>,
  "feedback": [
    "<feedback point 1>",
    "<feedback point 2>",
    "<feedback point 3>"
  ]
}
Resume: <extracted text>
The response from the model is parsed directly with JSON.parse() via result.output_text. The ats_scores value from the model response is stored as score in the ats_scores database table. If the model deviates from the expected format, the parse will throw and the upload transaction is rolled back.

​
Environment Variable

The GROK_API environment variable must be set to a valid Groq API key. Without it, the OpenAI client initialization in grok.js will fail and all resume uploads will return a 500 error at the AI analysis step.
# .env (backend)
GROK_API=gsk_xxxxxxxxxxxxxxxxxxxxxxxxxxxx
For the best results, tailor your resume to each job you apply for. Upload a version that includes keywords from the specific job description — the ATS model rewards relevance between your resume text and common job-posting language. A score above 75 is generally considered strong for automated screening systems.

Build docs developers (and LLMs) love

Get started for freeTalk to us