All Posts
How We Built an AI CV Screener That Does a Full Day of Hiring Work in 30 Minutes
Published
Category
Read Time
10 min
Author
Uzair Sayyed
Tech Lead
Screening 30 CVs manually takes most people a full working day. You open each file, read through it, try to remember what you were looking for, compare it against the last one you read, and make a judgement call - repeated 30 times, with your attention getting worse with every hour. We built a system that does that same screening in under 30 minutes. Consistent output every time. Structured results you can sort, filter, and export to a CSV. No expensive ATS software. No subscription. Just a well-engineered prompt and an AI chat window.This is the exact system. I'm going to walk you through it completely.
The Hiring Problem Nobody Talks About
The problem with CV screening isn't that it's hard. It's that it's exhausting and inconsistent. The fifth CV you read gets less attention than the first. The candidate you screen at 4pm gets a different standard applied than the one you screened at 10am. Human fatigue affects hiring decisions in ways nobody wants to admit.
For most small businesses, startups, and even mid-sized companies in India, there's no dedicated recruiter. The hiring sits with the founder, the department head, or an HR person managing ten other things simultaneously. They're reading CVs between meetings. That's not a process - that's a guess.
The obvious solution is an ATS. But enterprise ATS platforms are expensive, require setup, and are overkill if you're hiring for one or two roles a quarter. Most businesses in India don't need a full ATS. They need something that makes screening faster and more consistent without a long-term software commitment.
That's what we solved with a prompt-based AI screening system.
What We Actually Built and How It Works
The core idea is simple: instead of letting AI be a generic assistant, you train it through a master prompt to behave as a specialist CV screener for a specific role at a specific experience level.
The master prompt is not a generic "please review this CV" message. It's a structured brief that tells the AI exactly what role it's screening for, the required skills, tools and experience for that role, the experience bracket, the education requirements and which degrees are acceptable, red flags to look for and flag immediately, and the exact output format you want every time consistently.
You build one master prompt per role per experience bracket. A prompt for a junior ReactJS developer is different from a prompt for a mid-level one. The evaluation criteria, the acceptable gaps, the expectation on portfolio and projects - all of it shifts. The prompt needs to reflect that.
Once the prompt is live, you feed CVs into it. The AI evaluates each one against your criteria and returns a structured assessment. Same format, every time.
Takeaway: The quality of the output is entirely determined by the quality of the prompt. A vague prompt gives vague answers. A specific, well-structured prompt gives you a decision-ready assessment in seconds.
Step 1: Build the Master Prompt Per Role
This is where the real work happens - and you only do it once per role type.
What the Master Prompt Must Define
Role context: What does this person actually do day to day? Not just the job title - the real responsibilities. "ReactJS developer" tells the AI a category. "Builds client-facing web apps using ReactJS and NextJS, works directly with the design team, needs to be able to pick up Figma files and translate them to pixel-perfect responsive code" tells it what to actually look for.
Required skills vs good-to-have skills: Split these clearly. Required means a CV without it gets a No. Good-to-have means it earns bonus points but isn't disqualifying. If you blur this line in the prompt, the AI will blur it in its output.
Experience bracket: State the minimum and maximum years of experience acceptable. State what you expect someone at that level to have already done - not just titles, but actual output. A 2-year developer should have shipped live projects. A 4-year developer should have led at least part of a build.
Education requirements: Specify if a computer science degree is required, preferred, or irrelevant. For some roles, a strong portfolio from a self-taught developer is worth more than a degree from a mediocre college. Tell the AI what your standard actually is.
Red flags: What would immediately disqualify a candidate? Employment gaps over 12 months with no explanation? No live projects to show? Skills listed but not reflected in any work experience? Define these explicitly so the AI flags them consistently.
Output format: This is critical. Define exactly what you want the AI to return for each CV. Our format looks like this:
Candidate name : [name]
Shortlist : YES / NO
Strongest point : [one sentence]
Main concern : [one sentence]
Skills match : [score out of 10]
Experience match : [score out of 10]
Education match : [score out of 10]
Overall fit score : [score out of 10]
Recruiter note : [2 to 3 sentences of context]
Every CV comes back in this exact structure. You can read 30 assessments in 10 minutes and immediately know who to call.
If you want us to build a screening system like this for your next hiring round — or automate a different part of your business operations - this is exactly the kind of work we do at Nipralo - we've compressed months of manual work into hours using AI. Let's talk about what that looks like for you.
Step 2: Feed CVs Into the AI - Two Ways to Do It
This is where most people get stuck, because different AI tools handle file uploads differently - and some charge for it.
Route A: Pro Version - Direct PDF Upload
If you're using Claude Pro, ChatGPT Plus, or Google Gemini Advanced, you can upload PDF CVs directly into the chat. Paste your master prompt first, then upload the CVs one by one or in batches if the tool supports it.
This is the cleanest experience. The AI reads the PDF directly, applies your prompt, and returns the structured output. For a batch of 30 CVs, this takes about 20 to 25 minutes total.
Route B: Free Version - Text Extraction Using Microsoft PowerToys
If you're not on a paid plan - or you want to avoid that cost - here's the free alternative we actually use.
Microsoft PowerToys is a free utility for Windows built by Microsoft. It includes a Text Extractor tool that lets you copy any text visible on your screen with a keyboard shortcut - including text inside a PDF that would otherwise be locked or unselectable.
The workflow:
Microsoft PowerToys - it's free and takes under 2 minutes
Open the CV PDF on your screen
Use the PowerToys Text Extractor (default shortcut: Win + Shift + T) to highlight and copy the full text of the CV
Paste it directly into your AI chat window, after your master prompt
The AI reads the pasted text exactly as it would a direct file upload
The output quality is identical. The only difference is it takes an extra 30 seconds per CV to copy and paste. For 30 CVs, that's still well within the 30-minute window.
If you're on Mac or prefer a browser-based option, Smallpdf, ILovePDF and PDF2Go all offer free PDF-to-text conversion online. Copy the extracted text and paste it into your AI tool the same way.
Takeaway: You don't need a paid subscription to make this work. The system runs on any AI chat window - free or pro - as long as you can get the CV text in.
Step 3: Read the Output and Export to CSV
Once you've run all CVs through the AI, you have 30 structured assessments sitting in the chat. The next step is getting that data into a format you can sort and share.
Exporting to CSV
Ask the AI to compile all the results it just produced into a CSV table. A simple prompt at the end works:
"Please compile all the candidate assessments from this session into a CSV table with the following columns: Candidate Name, Shortlist (Yes/No), Skills Match Score, Experience Match Score, Education Match Score, Overall Fit Score, Strongest Point, Main Concern."
The AI returns a formatted CSV block. Copy it, paste it into any text editor, save it as .csv, and open it in Google Sheets or Microsoft Excel. You now have a sortable, filterable table of every candidate - ready to share with anyone involved in the final decision.
Sort by Overall Fit Score. Filter to show only Shortlist = YES. Forward the sheet to the hiring manager. Done.
What used to take a full working day of reading, noting, comparing, and compiling is now a 30-minute process with a clean, consistent output at the end.
Takeaway: The CSV export turns AI output into a shareable, decision-ready document. One person runs the screening, everyone sees the results - no manual summarising required.
What This Looks Like in a Real Hiring Round
We ran this system for a hiring process with 30+ CVs for a technical role. Before this system, the screening alone was taking most of a working day - reading through each CV, making notes, trying to maintain consistency across the batch.
With the master prompt system in place, the entire screening round ran in under 30 minutes. The CSV output was shared with the hiring decision-maker immediately. The shortlist was ready within the hour. First calls were scheduled the same afternoon.
The consistency improvement was just as significant as the time saving. Every candidate was evaluated against exactly the same criteria, in exactly the same order, with exactly the same output format. No fatigue, no variation, no "I think I remember this candidate was stronger but I can't quite recall why."
The system doesn't just save time. It makes better decisions possible by removing the variables that make manual screening unreliable.
What AI Screening Can't Replace
I want to be direct about this because it matters.
AI screening evaluates what's written in a CV against criteria you define. It is extremely good at that specific task. But it cannot assess tone in an interview, cultural fit, communication style, genuine motivation, or the instinct a good hiring manager develops over years of interviewing people.
The shortlist the AI produces is not a hiring decision. It's a qualified starting point. The humans on your team still do the interviews. They still make the call.
What changes is the quality and speed of what they're working from. Instead of 30 CVs with no structure, they're starting from a ranked shortlist with clear notes on every candidate's strengths and concerns. That's a much better use of human judgement than reading PDFs for eight hours.
Use AI for what it's genuinely good at - consistent, fast, criteria-based filtering. Keep humans in the decisions that actually need human judgement.
Conclusion
Screening CVs manually is one of the most time-consuming and least consistent parts of hiring. A well-built AI screening system - built on a master prompt that's specific to the role, the experience bracket, and your exact criteria turns a full day of work into 30 minutes of structured, exportable results.
We built this system and it's working. No expensive software, no ATS subscription, no elaborate setup. Just a prompt, a chat window, and a clear output format that makes the right hiring decisions faster.
If you want to automate a part of your business that's currently eating time your team doesn't have, WhatsApp us at +91 79770 28431 - we'll tell you in 20 minutes whether AI can compress it.
Frequently Asked Questions
Can I use ChatGPT to screen CVs?
Yes, and it works well when you give it a structured master prompt rather than a vague request. The key is defining the role requirements, experience bracket, red flags, and output format inside the prompt before submitting any CVs. With a properly built prompt, ChatGPT, Claude, or any capable AI can evaluate each CV against your criteria consistently and return a structured assessment in seconds. The quality of the screening depends entirely on the quality of the prompt.
How do I automate CV screening without expensive software?
You don't need an ATS or any paid recruitment software. Build a master prompt in any AI chat tool — free or paid — that specifies your role requirements, skills, experience expectations, and the output format you want. Then feed CVs into it by uploading PDFs directly on paid plans, or by copying and pasting CV text using a free tool like Microsoft PowerToys on Windows. At the end of the session, ask the AI to compile all results into a CSV table and export it to a spreadsheet.
What should a CV screening prompt include?
A good CV screening prompt should include: the role title and actual day-to-day responsibilities, required skills vs good-to-have skills clearly separated, the acceptable experience bracket with what you expect at that level, education requirements and whether alternatives are acceptable, specific red flags to flag and disqualify on, and the exact output format you want returned for every candidate. The more specific the prompt, the more consistent and useful the output.
How do I export AI screening results to a CSV or Excel file?
After processing all CVs in a session, ask the AI to compile all the candidate assessments into a CSV table with your chosen columns — typically name, shortlist yes or no, individual scores, strongest point, and main concern. The AI returns a formatted CSV block. Copy that text, paste it into a text editor, save the file with a .csv extension, and open it in Google Sheets or Microsoft Excel. You can then sort by overall score, filter to shortlisted candidates, and share the sheet directly with anyone involved in the hiring decision.
Is AI CV screening accurate enough to use for real hiring decisions?
AI screening is highly accurate at evaluating what's written in a CV against criteria you define — skills, experience, education, red flags. It applies the same standard to every candidate without fatigue or bias, which often makes it more consistent than manual human screening across a large batch. However, it cannot assess communication, cultural fit, or genuine motivation. The output should be treated as a qualified shortlist — a structured starting point — not a final hiring decision. Humans still conduct interviews and make the final call.
