AI Resume Screening in India

Accelerate recruitment with AI

Recruiting teams routinely receive hundreds of resumes for a single opening. Screening every one by hand costs weeks of a recruiter's time. unoblox scores and ranks candidates against a job description through one OpenAI-compatible endpoint, billed in rupees.

Why AI resume screening

Human screeners get tired and inconsistent by resume #200. A language model reads every resume with the same attention as the first — extracting skills, experience, and education, and matching them against your job description in seconds rather than minutes.

Recommended models

Model	Input ₹/1M	Output ₹/1M	Use case
Qwen3 235B-A22B	₹9.07	₹55.44	High-volume first pass
GPT-4.1	₹201.6	₹806.4	Skill and keyword extraction
Claude Opus	₹504	₹2520	Deep, contextual final-round review

Prices per 1M tokens.

Screening flow

from openai import OpenAI

client = OpenAI(
    api_key="ub-gw-...",
    base_url="https://api.unoblox.ai/v1"
)

job_description = "Python developer, 3+ years, AWS, Django"
resume_text = "... candidate resume text ..."

prompt = "Job: " + job_description + "\nResume: " + resume_text + "\n\nReturn JSON: {must_have_skills_present: bool, experience_match: str, concerns: [str], fit_score: int}"

response = client.chat.completions.create(
    model="openai/gpt-4-1",
    messages=[{"role": "user", "content": prompt}],
    response_format={"type": "json_object"}
)

print(response.choices[0].message.content)

A two-stage screen

1. Stage 1 — cheap pass (Qwen3 235B-A22B): every resume gets a fit score 1–10 against must-have criteria. Cuts the pile to the top 20–30%.
2. Stage 2 — deeper read (GPT-4.1 or Claude Opus): the shortlist gets a fuller extraction — project depth, tenure patterns, red flags — for the recruiter to review.
3. Human decision: a person makes every call to interview or reject; the model only orders the queue.

Keeping the process fair

• Ask the model to ignore name, age, gender, and photo when scoring.
• Standardize the prompt across every candidate for the same role — never hand-tune per resume.
• Periodically audit the shortlist for skew against any group, and adjust the prompt or criteria if you see it.

What it costs

At roughly 400 tokens per resume and GPT-4.1 input rates, screening 1,000 resumes costs about 1,000 × 400 × ₹201.6 ÷ 1,000,000 ≈ ₹81 for the extraction pass alone — cheap enough to run on every applicant rather than a sampled subset.

Use cases

• Startup hiring at high volume with a small recruiting team
• IT staff augmentation and bulk contractor screening
• MBA and graduate program admissions review
• Consulting-firm associate recruitment cycles

Frequently asked questions

Q: Can AI screening replace human judgment? A: No. Use it to rank and shortlist the top 20–30%; a person still conducts interviews and makes the final call.

Q: Does AI screening introduce bias? A: It can, if trained or prompted on biased historical hiring patterns. Mitigate by masking demographic details, auditing shortlist composition regularly, and keeping prompts identical across candidates.

Q: How do I handle resumes in PDF or image format? A: Use GPT-4o to read the image or PDF directly, or turn it into plain text first with an OCR step, then feed that text to your screening model.

Q: Can I screen for soft skills like communication? A: Yes, to a degree — include a cover letter or writing sample in the prompt and ask the model to infer tone and clarity, but treat this as a weak signal compared to a live interview.

Q: How much does screening 1,000 resumes cost? A: See the cost breakdown above — at GPT-4.1 rates, the extraction pass runs about ₹81 for 1,000 resumes, before any deeper second-stage review.

Q: Does this understand Indian degrees and certifications? A: Claude Opus and GPT-4.1 both recognize major Indian universities and common certifications (NIIT, Aptech, AICTE-affiliated institutes) — spell out anything obscure in your prompt to be safe.

Get started in rupees → https://unoblox.ai/sign-in