I will scrape websites, annotate images, and preprocess datasets
AI Developer Machine Learning Engineer Python FullStack Developer
About this Gig
Clean data decides how far a project gets. If records are inconsistent, folders are a mess, or images carry no labels, the algorithm is not the problem.
I collect, label, clean and prepare datasets:
- Web scraping: permitted public data into CSV, Excel or JSON, with a column dictionary
- Image annotation: classification labels, categories and consistent class definitions
- Bounding boxes: object detection annotation in YOLO or COCO format
- Preprocessing: duplicates, missing values, data types, formatting, normalisation
- Dataset preparation: folder structure, image-label pairing, splits, validation, documentation
Tools: Python, Pandas, BeautifulSoup, Requests, Selenium/Playwright, OpenCV. Outputs: CSV, Excel, JSON, YOLO, COCO JSON, organised folders.
Every delivery includes the processed files, a quality summary and notes on what changed. Originals are preserved.
Message me before ordering with your data source, quantity, classes and required output format so scope and price are agreed first. Public and permitted sources only. I do not train models or build applications as part of this gig.
My Portfolio
FAQ
Can you scrape any website?
No — and any seller who says yes is guessing. I collect from public sources you are entitled to use, respecting rate limits and terms. Sites behind logins, paywalls, CAPTCHAs or anti-bot protection are out of scope, and sites that block automated access may be declined after inspection.
Which data formats do you deliver?
CSV, Excel (XLSX), JSON, YOLO .txt files with a data.yaml, COCO JSON, organised image folders, and documentation in Markdown or PDF. One output format is included in a package; a second format for the same data is quoted as an extra.
Can you annotate images with bounding boxes?
Yes. I annotate objects against a convention agreed before the work starts: how tightly the box follows the object, whether partly visible objects count, how many boxes per image, and how classes are named. Boxes can be delivered with your class list applied exactly as written.
Do you support YOLO and COCO formats?
Yes, both — YOLO normalised .txt files with a data.yaml, and COCO JSON with images, annotations and categories. I can also convert between them, or produce a CSV of pixel coordinates, if your pipeline needs that.
Can you clean and preprocess an existing dataset?
Yes. That includes duplicate removal with a count, missing-value handling as you decide it, type and date correction, whitespace and case normalisation, column renaming, image resizing, and validation of the output. Originals are preserved and a short transformation log is delivered with the cleaned
Can you organise images and labels into folders?
Yes. I set up images/, labels/ and annotations/ structures, split into train/validation/test as agreed, check that every image has a matching label file, report orphan files, and reconcile counts across splits.
Do you train machine learning models?
Not in this gig. This service ends at the data: collection, labels, cleaning, structure and documentation. Model training, fine-tuning and evaluation are a separate service — a prepared dataset from here can feed that work, but you are not paying for it in this order.
Can you handle large datasets?
Within the quantities stated in each package (up to 1,500 records or 250 annotated images in Premium). Larger volumes are possible but are quoted separately after I see the sample, because annotation effort scales with images × classes × objects per image, not with file size alone.
How is annotation quality checked?
Against criteria agreed before the order: the class definition sheet, your written instructions, an agreed review percentage (for example 10% of images re-checked), and the conventions recorded in the brief. Ambiguous cases are logged rather than silently decided. What I do not offer is a blanket cl
Do I need to send anything before ordering?
A short brief: what the data is for, where it lives, how much of it there is, the class names if annotation is involved, and the format you need. Screenshots of an existing annotation file help a lot. Send it as a message and you will get a scope note with quantity, format, price and delivery time.

