I run FirstDay AI, a paid job-search service in India. The tool in this post is free and MIT licensed. These are our rules, not Naukri's ranking algorithm, which isn't public. We're not affiliated with Naukri.

If you've looked for a job in India, you've probably had this experience: you apply to a hundred roles on Naukri, update your profile every morning, and nobody calls.

At FirstDay AI we run job searches for people. We apply on their behalf and follow up with HR. Before any of that, we fix their Naukri profile, because a lot of the calls don't come from applications at all. They come from recruiters searching the candidate database.

After doing that by hand again and again, the fixes started to look like rules. So we wrote them down as code, put them on GitHub, and gave them a score out of 100. This post walks through what those rules check and why.

Repo: https://github.com/FirstDay-AI/naukri-profile-optimizer

naukri-optimizer check scoring a sample Data Analyst profile 17/100, then the fixed profile 100/100

How recruiter search works, in one paragraph

A recruiter with an opening types a job title and a few skills into the search box. Naukri's own guidance says recruiters search for and shortlist candidates based on their skills, and that the resume headline helps them scan whether you fit a role. Then they narrow the list with filters: location, notice period, salary, experience and how recently the profile was active. Then they read the top of each profile for a few seconds and decide whether to open it.

That gives three ways to fail: you don't match the search, you get filtered out, or you get skipped when they read. The scorer has checks for each.

The scoring model

Section Points What it checks
Headline 25 Length, target role named, role keywords, a number, filler words
Key skills 25 12 to 25 skills, core skills for your role, share of soft skills
Profile summary 20 Length, role keywords, a number, filler phrases
Employment 10 Designation matches target role, resume attached
Activity 10 Updated in the last 7 days
Career profile 10 Preferred locations, notice period, expected salary

Headline and key skills carry half the score, because they're what search matches on most directly.

Rule 1: name the role, then prove it

pts = 6 if is_title(n_head) else 0 add("headline_role", "Headline", pts, 6, "Headline names your target role." if pts else "Headline doesn't name your target role.", f"Start the headline with \"{role.get('headline_name', role['name'])}\" so recruiter searches match it.") 

is_title checks the headline against a list of titles recruiters use for that role. For Data Analyst, that list includes "data analyst", "BI analyst", "MIS analyst" and "reporting analyst".

The most common headline we see is some version of "Hardworking professional seeking a challenging position in a dynamic organisation". It matches no search. Compare it with:

Data Analyst | 3+ Years | SQL, Power BI, Advanced Excel, Python | [one result with a number] 

That's the template the tool generates. It puts the role first, then years, then 3 to 5 skills you actually have, then one result. The headline check also gives points for 3 or more role keywords, for including any number, and for avoiding filler. Length matters too: 60 to 250 characters gets full marks.

Rule 2: key skills are search terms, not a personality test

n = len(skills) pts = 5 if 12 <= n <= 25 else (3 if 8 <= n <= 11 or 26 <= n <= 40 else 0) ... matched_core = [k for k in role["core"] if has_kw(n_skills, k)] target = min(8, len(role["core"])) pts = round_half_up(15 * min(len(matched_core), target) / target) ... generic = [s for s in skills if s.lower() in GENERIC_SKILLS] share = (len(generic) / n) if n else 0 pts = 5 if n and share <= 0.2 else (2 if n else 0) 

Three checks:

  1. Count. 12 to 25 skills. With too few, you match fewer searches. With 60, the real ones get buried and the list looks copied.
  2. Core coverage. Each role has 10 core keywords. You get full marks for 8 of them. You don't need all ten, because no real person has every tool.
  3. Soft-skill share. If more than 20% of your skills are things like "Communication", "MS Office", "Teamwork" or "Time management", you lose points. Recruiters rarely type those into a search box. They're not bad qualities. They're just not search terms.

Keywords live in a JSON file with aliases:

{ "id": "data-analyst", "core": ["SQL", "Excel", "Power BI|powerbi", "Tableau", "Python", "Data Visualization|data visualisation", "Dashboards|dashboard", "Data Analysis", "Statistics", "Advanced Excel"] } 

Power BI|powerbi means "show Power BI, but also count powerbi". Matching runs on normalised text padded with spaces, so "Java" doesn't match inside "JavaScript":

def norm(text): t = (text or "").lower() t = re.sub(r"[^a-z0-9+#.]+", " ", t) t = re.sub(r"\.(?=\s|$)", " ", t) return " " + re.sub(r"\s+", " ", t).strip() + " " 

The character class keeps +, # and . so that C++, C# and Node.js survive normalisation.

Rule 3: the summary needs a number

pts = 5 if re.search(r"\d", summary) else 0 add("summary_metrics", "Profile summary", pts, 5, "Summary includes numbers." if pts else "Summary has no numbers.", "Add results: team size, % improvement, revenue, tickets per day, projects shipped.") 

It's the bluntest rule in the file, and it's the one people most often fail. A number turns "worked on reports" into something like "built [N] weekly MIS reports for a [N]-person sales team". It gives a recruiter something to ask about on the call.

The summary also gets checked for length (300 to 1,000 characters, about 3 to 5 sentences), for at least 6 role keywords used naturally, and for filler. The filler list is short and specific (abridged here):

FLUFF = ["hardworking", "seeking a challenging", "dynamic organization", "team player", "self motivated", "result oriented", "quick learner", "passionate", "dedicated", ...] 

Rule 4: freshness and filters

pts = 10 if updated <= 7 else (5 if updated <= 30 else 0) 

Recruiters can filter for recently active profiles, so a profile untouched for two months can fall out of those searches. The fix costs two minutes a week: add a skill, reword a line, re-upload your resume.

The career-profile checks are about filters too. If notice period, preferred locations or expected salary is empty, a recruiter who filters on that field may never see you. Notice periods of 30 days or less score highest. That's not a judgement. Shorter notice periods are simply what many openings ask for. If yours is 90 days, the tool suggests mentioning buyout or your last working day.

What the output looks like

Here's the included example profile, which is illustrative and not a real person, before any changes:

$ naukri-optimizer check examples/data_analyst_before.json Naukri profile score: 17/100 (Needs work) Target role: Data Analyst Headline [####----------------] 5/25 Key skills [#####---------------] 6/25 Profile summary [#-------------------] 1/20 Employment [--------------------] 0/10 Activity [--------------------] 0/10 Career profile [##########----------] 5/10 Fix these first: -11 pts 2 of 10 core Data Analyst skills listed. ... -10 pts Profile last updated 64 day(s) ago. ... - 7 pts Headline has 0 role keyword(s). ... 

Fixes are sorted by points lost, so the first line is always the biggest win.

The AI rewrite, and why it's bring-your-own-key

Scoring needs no network and no key. The optional rewrite command sends your profile and the score report to an AI provider you choose: OpenAI, Anthropic, any OpenAI-compatible API such as Groq or OpenRouter, or Ollama on your own machine.

The prompt has one rule that matters more than the rest: use only facts found in the profile, and never invent employers, numbers, degrees, tools or skills. Where a number would help but you didn't give one, it writes [X%] for you to fill in. It adds a missing core skill only if your profile text shows you've used it. An AI that makes up a Kafka project for you is a liability in the interview.

export ANTHROPIC_API_KEY=sk-ant-... naukri-optimizer rewrite profile.json --provider anthropic --model <model-name> 

Running it

pip install git+https://github.com/FirstDay-AI/naukri-profile-optimizer naukri-optimizer init profile.json # fill in your Naukri sections naukri-optimizer check profile.json 

There's also a JS port (js/scorer.js) that runs in the browser. A parity test in CI checks that the Python and JS versions give identical results, so the web version can't drift from the CLI.

If you'd rather not use a terminal, we host the same scorer at firstdayai.in/tools/naukri-profile-optimiser. The score is free and needs no sign-up, and your resume is read in the browser. The AI rewrite there asks for your name and WhatsApp number, because that's how we meet people who might want our paid service. Our team may contact you, so use the CLI if you'd rather not.

What I'd like from you

The rules are opinions written as code. Some are probably wrong for some roles. If you hire on Naukri, or you've been on either side of a recruiter search, I'd like to hear:

  • Which rule is wrong, and what you'd change it to
  • Which role is missing (adding one is a single JSON edit, see CONTRIBUTING.md)
  • Which keywords are outdated for 2026

Issues and PRs: https://github.com/FirstDay-AI/naukri-profile-optimizer

Disclosure: I run FirstDay AI, a paid job-search service in Chennai. The tool is free, MIT licensed and works without us. Not affiliated with Naukri.com or Info Edge (India) Ltd.