DEV Community

Joan Corbi
Joan Corbi

Posted on

How to win freelance automation clients: 3 working demos on GitHub

The client problem: "I do this every day by hand, can you automate it?"

That is the real freelance brief. Not "build me a SaaS", not "write me a bot".
It is: I open this page every morning, copy the prices into Excel, and it takes me half an hour.

Three years of doing this taught me one thing: you win the client with a working demo, not a proposal PDF. So I built three demos that answer exactly those three briefs — and published every one of them with the real output files committed to the repo.

This post is a walkthrough of the three. Each repo runs in under a minute after pip install -r requirements.txt.

1. "Can a computer do what my employee does in the browser?"

Repo: rpa-ecommerce-bot

The manual job: open a shop, read each product card, type the name and price into a spreadsheet. 30 minutes a day, with typos, with no evidence, with no way to audit it.

The repo does the same walk with Selenium + Chrome:

def collect_catalog(driver, limit):
    wait.until(EC.presence_of_element_located((By.CLASS_NAME, "product_pod")))
    rows = []
    for card in driver.find_elements(By.CLASS_NAME, "product_pod")[:limit]:
        name = card.find_element(By.CSS_SELECTOR, "h3 a").get_attribute("title")
        price = card.find_element(By.CLASS_NAME, "price_color").text.strip()
        ...
Enter fullscreen mode Exit fullscreen mode

Outputs: precios_YYYY-MM-DD.csv (UTF-8 BOM, opens clean in Excel) and a PNG screenshot as evidence. The repo commits both from a real run, and a selector-guia.md that teaches how to find the right selectors for your site with DevTools — because that is the only part that changes between clients.

The lesson that surprised me: I had written the script against a tutorial. The tutorial's search box did not exist on the live site. The selectors in the repo are verified against the actual HTML, not copied. Clients don't buy "it works in my demo". They buy "it works in your site" — and that means someone actually looked at the real DOM.

2. "Send me the whole catalog, it's 50 pages"

Repo: web-scraper-catalogo

Copy-pasting 1,000 products from 50 pages takes hours and never comes out complete. requests + BeautifulSoup follows the real pagination (<li class="next">) and extracts name, price, rating and availability:

$ python scraper.py --pages 5
OK  100 productos (5 paginas)
     en stock: 100  |  precio medio: 34.56
CSV   output/catalogo.csv
XLSX  output/catalogo.xlsx
Enter fullscreen mode Exit fullscreen mode

Two details that save you a client call:

  • Prices come out as numbers, not text. 51.77 stays a number, so the client can SUM, AVERAGE, build a pivot. A CSV full of 51.77 strings is a CSV you have to clean again — you just moved the manual work.
  • The .xlsx ships with filters and formatted columns. It is the difference between "here is a file" and "here is a file you can use today".

3. "My client sends me a table in an email or a web report, and I need it in Excel"

Repo: excel-from-html

This is the one I was not sure a repo could sell. Then I realised: this is the most common actual task in an operations team. The page only offers "export as HTML" or a screenshot.

html2excel.py takes a local file or a URL, extracts the table, and does the conversion that people do by hand for ten minutes:

In the HTML In the Excel Type
12.450,00 12450.0 number
45.581,25 45581.25 number
51,8% 0.518 (as %) number

European number formats are the trap: 12.450,00 means twelve thousand, not twelve point four five zero. to_number() handles both 1.234,56 and 1234.56 and 50%, and leaves genuine text as text. The committed muestra-resultado.xlsx in the repo is the actual output of a run — open it and check the cell types yourself.

Why demos > proposals

A proposal says "I can do this". A committed CSV from a real run shows it. Every repo in github.com/facturarapida follows the same contract:

  1. The manual problem, in one paragraph, in the client's words.
  2. The code, short enough to read in five minutes.
  3. The real output files from an actual execution, committed.
  4. How to adapt it to your site — because the client knows their own DOM better than I do, and the honest answer is "change three lines, here is how".

That last point is what makes it a service instead of a tutorial. A tutorial teaches you to build a scraper. A service takes your task, makes a demo with your type of data, and hands you the script plus the Excel.

What this is not

No scraping of sites that forbid it. No paywalled content. No "unlimited" anything. If a site's terms say no, the answer is a polite no and a different route. Every demo runs against a public sandbox site that exists to be scraped — the same way a client would approve a first test before touching production.

Try it

Pick the repo that matches your manual task, clone it, run it, and look at the output files. If it is close to what you need, the conversation with me is not "are you able to do this" — it is "how do I point it at my site".

That is the whole pitch.


All three repos are MIT-licensed. Portfolio.

Top comments (0)