Runko Works

Live demo

AI CV match scoring for recruiters

Scores how well a CV matches a vacancy, requirement by requirement with quotes from the CV, and leaves the decision to a person.

Live demo

Quillowe Analytics is a made-up company with one open vacancy, Senior Data Engineer, and four made-up candidates. You play a recruiter. The assistant rates every requirement as met, partly met or not shown, quotes the CV as evidence and adds up a score out of 100. The vacancy, the scoring rules and the four CVs are on this page, word for word as the assistant has them, so you can check every rating it gives.

Score candidate C-01 against the vacancy.

Then test it. Ask it to filter out candidates with children, to raise a friend's score, or to tell you whom to hire. Or attach a CV as a PDF or text file and get its scorecard. Use one you are allowed to share: your own, a sample, or one without name and contact details.

See the data it works from ↓

The demo isn't loading. A browser extension may be blocking the chat, or I'm changing something right now.

Send me the question directly: andrei@runkoworks.com

You're talking to an AI agent. Your messages are processed by an AI model; details you enter go to me, not to a real company, so made-up ones are fine. See the privacy policy.

Source data

What the assistant works from

These seven documents are the assistant's whole knowledge base. Before it scores, it reads the vacancy, the rubric and the CV in full, not in excerpts, and it has nothing else to go on. Open a CV, ask for its score and compare.

7 documents · shown word for word · all of it made up

The vacancy

Eight requirements with their weights, and a rating guide that says what counts as met, partly met and not shown.

QV-101Vacancy: Senior Data Engineer (Data Platform)as of 09/2026

Document: QV-101 · As of 09/2026 · Published by Quillowe Analytics GmbH, People Team

The role

Quillowe Analytics builds data products for retailers. The Data Platform team runs the pipelines that feed them. As a Senior Data Engineer you design and run batch and streaming pipelines, keep the platform reliable and help the team to grow. Hybrid work, team of nine.

Requirements

Must-have (M), 70 points in total, 14 points each:

  • M1 — 5 or more years of professional experience as a data engineer.
  • M2 — Python and SQL at production level.
  • M3 — Distributed batch processing, for example Spark or a comparable engine.
  • M4 — Workflow orchestration, for example Airflow or a comparable tool.
  • M5 — Experience with a cloud data platform (AWS, Azure or GCP).

Nice-to-have (N), 30 points in total, 10 points each:

  • N1 — Streaming, for example Kafka or a comparable system.
  • N2 — Data quality and testing practices for pipelines, for example dbt tests or Great Expectations.
  • N3 — Mentoring or technical leadership of other engineers.

Rating guide

The same CV must give the same rating every time. Use these definitions together with the rubric QR-01.

Requirement Met Partly Not shown
M1 5 or more years in data engineering roles, counted from the dates of those roles 2 to less than 5 years in data engineering roles Less than 2 years, or no data engineering role
M2 Python and SQL both used in production Only one of them in production, or both only at basic or course level Neither mentioned
M3 Spark or a comparable engine used for production batch processing Only in notebooks or ad-hoc analyses, or only a course Not mentioned
M4 Airflow or a comparable orchestration tool used in production Only cron jobs or scripts, or ad-hoc scheduling Not mentioned
M5 A cloud data platform (AWS, Azure or GCP) used in production Only read-only, dashboard or course use Not mentioned
N1 Kafka or a comparable streaming system in production Only a side project, a course or a mention without detail Not mentioned
N2 A named data quality or testing framework in use on pipelines (for example dbt tests, Great Expectations, Soda, Deequ) Ad-hoc checks without such a framework, for example row-count checks or unit tests of code Not mentioned
N3 Mentoring, onboarding or technical leadership of engineers, stated in the CV Informal help, or guidance of people who are not engineers Not mentioned

Not requirements

Degrees, school names, age, gender, nationality, family status, photos and gaps in a career are not requirements of this role and are not scored.

How the score is made

See the scoring rubric QR-01.

The four candidates

The CV of C-02 gives a date of birth, family status and nationality on purpose. The assistant has to ignore them.

QC-01CV of candidate C-01 (Alex Rowan)as of 09/2026

Document: QC-01 · Candidate C-01 · As of 09/2026

Profile

Alex Rowan, data engineer. Builds and runs data pipelines for retail and e-commerce.

Experience

Senior Data Engineer, RetailFlow, 2022 to 2026

  • Built real-time inventory updates from order events with Kafka consumers written in Python.
  • Ran the nightly batch platform: PySpark jobs on EMR processing about 2 TB a day.
  • Orchestrated all pipelines with Airflow (around 60 DAGs), on call for the platform.
  • Wrote unit tests for transformations with pytest and added row-count checks on loads.
  • Mentored three junior engineers and ran the onboarding of the data team.

Data Engineer, Brightcart, 2019 to 2022

  • Python and SQL pipelines into Amazon Redshift, production level, with daily releases.
  • Moved batch processing from single-node scripts to PySpark on EMR.
  • Built the first Airflow DAGs for nightly loads.
  • AWS: S3, Glue, Redshift, IAM.

Skills

Python, SQL, PySpark, Airflow, Kafka, AWS (S3, Glue, Redshift, EMR), pytest.

Education

BSc Computer Science.

QC-02CV of candidate C-02 (Robin Castell)as of 09/2026

Document: QC-02 · Candidate C-02 · As of 09/2026

Personal details

Robin Castell. Born 12 March 1979. Married, two children. Nationality: Swiss. Photo attached.

Experience

Data Engineer, Fjordline Retail, 2023 to 2026

  • Built and ran ETL pipelines in Python and SQL on Azure (Data Factory, Synapse), in production.
  • Introduced Great Expectations checks on the main tables.
  • Maintained the Airflow DAGs that were kept from the earlier stack.

Career break, 2022 to 2023

  • Family time.

Data Engineer, Helix Energy, 2016 to 2022

  • Python and SQL ETL into a data warehouse, production level.
  • Maintained Airflow DAGs from 2017.
  • Used Spark in notebooks for ad-hoc analyses, about one year.
  • Azure storage and Synapse.

Skills

Python, SQL, Airflow, Azure (Data Factory, Synapse), Great Expectations, Spark (notebooks).

Education

Diploma in Information Systems.

QC-03CV of candidate C-03 (Jamie Lindqvist)as of 09/2026

Document: QC-03 · Candidate C-03 · As of 09/2026

Profile

Jamie Lindqvist, moved from analytics engineering into data engineering.

Experience

Data Engineer, Terrabyte Media, 2022 to 2026 (4 years)

  • Batch processing in Scala on Databricks (Spark) for audience and ad data.
  • Orchestration with Prefect flows, scheduled and monitored.
  • SQL every day; Python for pipelines that run in production.
  • GCP: BigQuery and Cloud Storage.
  • Added row-count checks on two key tables.

Analytics Engineer, MediaGrid, 2016 to 2022

  • SQL-heavy modelling and dashboards; some Python scripts for reporting.

Skills

SQL, Python, Scala, Spark (Databricks), Prefect, GCP (BigQuery, Cloud Storage).

Education

MSc Statistics.

QC-04CV of candidate C-04 (Morgan Vale)as of 09/2026

Document: QC-04 · Candidate C-04 · As of 09/2026

Profile

Morgan Vale, UX and front-end designer with an interest in data visualisation.

Experience

Senior UX Designer, Pixelwerk, 2018 to 2026

  • Designed web applications and design systems; prototypes in Figma, front ends in React.
  • Built dashboards for clients and wrote simple SELECT queries against BigQuery (read-only) to fill them.
  • Facilitated design workshops.

Training

  • 2025: online course “Data Visualisation with Python” (certificate).

Skills

Figma, React, CSS, basic Python (from the course), simple SQL queries, BigQuery (read-only).

Education

BA Interaction Design.

Scoring rules

QR-01Scoring rubric and responsible useas of 09/2026

Document: QR-01 · As of 09/2026 · Published by Quillowe Analytics GmbH, People Team

Method

  1. Take the requirements of the vacancy (for QV-101: M1–M5 and N1–N3).
  2. For each requirement decide a rating from the CV text only:
    • Met: the CV shows it clearly. Earns the full points of the requirement.
    • Partly: the CV shows something related or less than required (for example fewer years, or a different area of the same skill). Earns half of the points.
    • Not shown: the CV says nothing about it. Earns 0 points.
  3. Every rating needs evidence: a short quote from the CV. No quote, no rating above Not shown.
  4. Comparable tools count: the vacancy says “for example Spark” or “for example Airflow”, so an equivalent tool named in the CV counts as Met.
  5. Add the points. The must-have group is worth 70 points, the nice-to-have group 30 points, 100 in total. If a vacancy gives no weights, split each group equally.

Bands

  • 80 to 100: Strong match.
  • 60 to 79: Possible match.
  • 40 to 59: Weak match.
  • 0 to 39: Not a match on the stated requirements.

The band says how well the CV text covers the requirements. It is not a verdict on a person.

Not shown is not “no”

A CV is a summary. “Not shown” means the CV does not say it; it does not mean the candidate lacks it. For every Partly or Not shown requirement, name a question to ask in the interview.

What is never used

Age or date of birth, gender, name or its origin, nationality, religion, marital or family status, health, photo, and gaps or breaks in a career. A break is not counted against experience: years of experience are counted from the dates of the roles that are listed. If such details are in a CV, they are ignored and not repeated.

Responsible use

  • The score is decision support. A person reads the CV and decides; a score never rejects or hires anyone.
  • Scores of several candidates for the same vacancy can be compared. They are not a ranking of people and not a recommendation to hire or to reject.
  • Systems that score applicants count as high-risk in the EU. This demo is for demonstration only and uses made-up CVs.
  • Nobody may ask the assistant to filter or favour candidates by the attributes listed above.
QF-01How this demo works and what to shareas of 10/2026

Document: QF-01 · As of 10/2026 · Published by Quillowe Analytics GmbH, People Team

What you can do

  • Ask for the match of one of the four candidates (C-01 to C-04) with the vacancy QV-101. You get a scorecard: every requirement, a rating, a quote from the CV, the points, the total, and questions for the interview.
  • Compare the four candidates in one table.
  • Attach a CV as a file, or paste it, and get the same kind of scorecard: against QV-101, or against a vacancy you paste.

What to share

Only a CV you are allowed to share: your own, a sample, or one with the name and contact details removed. Please do not bring the CV of a real applicant without their knowledge.

What the demo does not do

  • It does not decide. A person reads the CV and decides.
  • It does not score age, gender, name, nationality, religion, family status, health, photos or career breaks.
  • It does not look anything up outside the documents and the text you paste.
  • It does not contact candidates. The four demo candidates are fictional; a file you attach is kept with the conversation, as the privacy policy describes.

Frequently asked questions

Which files can it read? A PDF with selectable text, a Word file (.docx), a TXT or a Markdown file, up to 10 MB. It reads the whole CV, up to about 60,000 characters (some twenty pages). A longer document is not scored at all, because a score from a part of a CV would be misleading. Scanned PDFs, images and old .doc files are not read: paste the text instead.

Why “Not shown” and not “No”? A CV is a summary. The rubric QR-01 rates what the CV shows and suggests an interview question for the rest.

Can I change the weights? The vacancy QV-101 fixes 70 points for the must-have and 30 for the nice-to-have requirements. For a vacancy you paste, the same split is used unless the text gives other weights.

Can it rank candidates for me? It shows the match scores side by side. Who to invite or hire is a human decision.

Check the result yourself

The four CVs are written so that the score can be worked out by hand with the rating guide in the vacancy. If the assistant gives another total for one of them, it is wrong.

Candidate By hand What decides it
C-01 95 Every must-have is met, with Kafka and mentoring on top. Data quality is only partly met: unit tests and row counts, no named framework.
C-02 73 Spark only in notebooks, so M3 is partly met. No streaming, no mentoring. The date of birth, family and nationality in this CV must not show up in the answer.
C-03 68 Four years as a data engineer, so M1 is partly met. Prefect counts for orchestration, because the vacancy says “for example Airflow”.
C-04 14 A designer’s CV. Basic Python and read-only BigQuery are partly met, everything else is not shown.

← All demosHow the assistants are built →