EN TR
Get a quote
View all services
Web software

One app carries the brochure pages and the request queue.

System architecture

The pieces and their boundaries are settled before code.

Enterprise & commercial software

Several institutions, one system, records kept apart.

Mobile app

One codebase, both stores: iOS and Android.

Software consulting

We take a running system over and improve it.

AI consulting

AI only where the problem actually calls for it.

Get a quote
EN TR

Process Tracking · Kartal

Process Tracking for Kartal businesses

A single printed copy stops a process where it stands: turning what is inside it into records means somebody typing item by item. In a workshop around Soganlik or Yakacik in Kartal, or on a service floor near the health campus, that first link is where the chain can jam. When you go looking for process tracking software the blockage is rarely reporting; it is the input. In the tool we wrote for Vipkurum a booklet page is split into individual questions, and image processing draws the boundary.

Get a quote Tell us about your project

Process Tracking — Kartal

Kartal's texture is manufacturing along the E-5 and the ground around the health campus. If work begins with a measurement, a check or a printed booklet, the first record starts on paper. The work closest to this district's sector profile is Vipkurum's question cropping tool, written for education technology. Where it does not overlap: a production measurement form, an institutional file or a staff roster are outside what we have built, and we promise nothing there. What we describe is narrow and concrete — how one printed page turns into individual records. Python and OpenCV do the cropping, Gemini and OpenAI models handle the matching, and the cropped images and records sit in MongoDB. We keep no office; the work runs remotely.

  • Image processing decides where a question starts and ends: OpenCV draws the boundary, and that call is never left to a language model's reading.
  • The answer key stops being a separate document, because the model writes the matching answer into the cropped question's own record.
  • A question nobody has can be written from existing examples together with its answer, and the cropped images and records meet in one pool.

Why Algoriel

  • Keeps working offline

    The field app keeps recording with no connection, and the queue reaches the server in order once signal returns. That is how AlgoKlima runs.

  • Entry from QR and location

    Where no fixed terminal can be installed, entry and exit are logged from a QR code and a location reading, and entitlement is worked out from those readings.

  • Records land the same hour

    Jobs close in the field, so the office sees where each one stands without waiting for end of day.

  • Office and field read one source

    The office panel and the field app read the same information, so nobody keeps a spreadsheet alongside it.

What we build

Process Tracking

  • What the record attaches to

    Rather than mapping a process in the abstract, we decide what a record hangs on. On the construction system a row belongs to a person and a site at once, and the two are never separated. In AlgoRandevu we are building a structure where a booking can attach to a person, a room or a piece of equipment.

  • The entry that stays open

    Instead of a live dashboard, one rule: a matched check-in and check-out produce the hours, and a check-in with no scan out stays open. Whatever is stuck shows up as that gap.

  • Who was where, on the record

    Because a single row carries both the person and the site, totals separate by site. When one worker spends the month across two sites, those hours never fall into the same bucket.

  • What is left when the figure is disputed

    Because payroll is derived from matched pairs, the number argued over at month end has a time- and location-stamped record behind it. There is no parallel calculation sheet to reconcile against; the dispute happens over the record itself.

  • Records taken beyond coverage

    In AlgoKlima's service flow the chain does not break where the network does. Once the connection returns the waiting entries are processed in the order they were created, and an edit made in the field and one made in the office merge into a single record through conflict resolution.

  • Process rules kept out of the code

    Rules like the length of a step and the gap between consecutive ones belong in a product setting, not in the code. The aim is that changing one does not mean shipping a release, and that is what we are building into AlgoRandevu right now.

Work like this

Process Tracking · Case study

Vipkurum question cropping — OpenCV splits the page, the key enters the record

The cropping and key matching tool we wrote under the Vipkurum platform. A booklet page is separated into individual questions by image processing: OpenCV works out the boundaries, never a language model. Gemini and OpenAI models attach the answer from the key to the question it belongs to, so the key stops being a separate document and enters that question's record. A question nobody has can be written from existing examples, with its answer. Cropped images and records sit in MongoDB, reached through Node.js from the content management side on Laravel. Cropping stays in image processing, interpretation in the model.

Read the case study →

Working with Kartal teams, remotely

We have no office, so work for this city runs remotely and we do not offer on-site support. What was done, and why it was done that way, stays in writing. The project pages mark which builds came out of a partnership.

  • The first release ships as a working piece; extras follow later.
  • Every client we know of is in Turkey.
  • Decisions stay in writing; progress shows in working screens.

Frequently asked

  • Our paper is an inspection form, not an exam. What does this tool tell us?

    It was written for exam content and we say so plainly. We have no production form or institutional record work to show. One method is what carries: the decision about where to cut a page is made by image processing, while interpretation stays in a separate layer.

  • Does the AI do the cropping as well?

    No. OpenCV works out where a question begins and ends, computed from the image itself. The language model is only involved in matching the key and in writing new questions from an example. We kept those two jobs apart on purpose.

  • What if a match is wrong? Do you publish an accuracy rate?

    We publish no rate, because we have measured none. Whichever question's record the key was written into is where it shows, so a wrong match is corrected on the record itself rather than against a separate document.

  • Where do the cropped questions live, and what happens next?

    The images and records sit in MongoDB. The Python image processing service is called through Node.js, and the records attach to the content management side on Laravel. The pool becomes a queryable set of records rather than a stack of folders.

Process Tracking in other cities

The other cities where we publish this service are below, grouped by province.

Other services for this location

What stands behind your month-end figure today?

Get a quote or open the service