Get a quote

AI Consulting · Pendik

AI Consulting for Pendik businesses

A control sheet does not live long in manufacturing. A new model reaches the line and a row is added, a buyer asks for an item of their own, and the sheet comes off the office printer again the same day. In a small manufacturer or a multi-site service organisation around Yesilbaglar and Cinardere that rhythm may be familiar. Tie the reading step to a device and the rhythm breaks right there, because the device wants its own paper and its own layout.

AI Consulting — Pendik

Pendik leans on manufacturing SMEs and multi-site service organisations. If your sheet changes as the product changes, or every site wants a line of its own added, the form gets reissued mid-season. The piece of our portfolio sitting nearest that fabric is Vipkurum's optical mark reading tool, which came out of education technology. The gap deserves naming: production planning, a quality management system and service call handling fall outside it, and nothing gets promised where nothing can be demonstrated. A fresh layout is taken up on the Python side while OpenCV lifts the marks out; the reading service runs behind Node.js, and the exam record itself is kept by Laravel on MySQL.

  • A refreshed layout is absorbed on the Python side, so neither a reader nor device-specific paper joins the purchase list.
  • A sheet coming off the office printer that same morning holds nothing up, because the reading step waits on no special print stock.
  • We put the boundary in writing: exam marking is where this came from, so production planning and service call handling are not ours to claim.

Why Algoriel

  • The image does the reading

    Vipkurum reads optical form marks with OpenCV from the image of the sheet, not through the optics of a reader appliance.

  • Cropping apart from judgement

    Image processing decides where a question begins and ends; the language model only binds the answer to it. The two jobs never mix.

  • Model output lands in a record

    Vipkurum writes the answer key into the cropped question's own record instead of keeping it as a separate document.

  • A person has the last word

    At Tahlil.com the model drafts the reading and a doctor approves what reaches the patient; an unapproved reading never appears on screen.

What we build

AI Consulting
  • Where no model is needed

    Marks on an answer sheet are read from the form image itself; the threshold that decides whether a bubble counts as filled sits in the code, and no language model runs in that step.

  • Splitting the job in two

    Computer vision sets the question boundaries; a language model then ties each answer in the key to its question. Cropping accuracy never rests on a model's interpretation.

  • Removing the hardware step

    Marking stops depending on a reader device; when the form layout changes, we adjust the reading service in software instead of ordering new hardware or device-specific forms.

  • Turning existing material into a pool

    A booklet page is split into individual questions, the answer key is written onto each question's own record, and the questions already on hand become the pattern for new ones.

  • Tying the output to a person

    At Tahlil.com AI prepares the first reading, a specialist physician writes and approves the interpretation, and responsibility for the text that reaches the patient stays with the physician who signed it.

  • Queueing the generation

    In Secretum text and images come out of two separate generation paths; image generation is not instant, so that work goes onto a queue, and the finished file is stored once instead of being regenerated in the feed. Secretum is the product we are building now.

Work like this

AI Consulting · Case study

Vipkurum optical mark reading — a new layout lands in code, not on a purchase list

Our optical mark reading work sits inside Vipkurum's education platform. Python and OpenCV lift the marks off the sheet image. In a manufacturing setting the consequence lands on the calendar: a fresh sheet layout becomes an adjustment in the code and takes effect the same morning, with no reader on order and no device-specific paper stock to wait for. A sheet arrives as a scan, Node.js fronts the reading service, and the outcome is stored by Laravel on MySQL. The boundary rides along: exam marking is where this came from, so production planning and service call handling are not covered.

Read the case study →

Working with Pendik 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 sheet changes with every new model. Do we have to come back to you each time?

    The adjustment happens in the image analysis layer. Once it is written down where each field falls on the paper, the rule goes into the code, and you are not waiting on a supplier's calendar in between.

  • How does this work without buying a device?

    The reading step lives in software and the sheet reaches the system from a scanner. Because no device is bought, the layout of the form is not squeezed into the dimensions one machine accepts and can stay shaped around your own work.

  • The paper we handle is an end-of-line control form, not an exam answer sheet.

    That tool came out of exam marking and we do not blur the difference. What carries across is where a layout gets met: change the form and what changes is an adjustment in the code. Quality management and production planning sit outside what we can demonstrate.

  • We run several sites and each one's form differs from the next.

    A layout difference is met as an adjustment on the software side, so no site needs a device of its own or a paper order of its own. We have no premises anywhere; every site is reached the same way, over the network.

Let us look at which step needs a model and which one needs computer vision.