Goombasys

Build the AI feature.Prepare the data behind it.

Engineers and annotators in Dhaka, building AI features and labelling training data for companies in the US and Europe.

Discuss your projectSee the two services
Our own drone footage of the Dhaka metro. Each object is segmented every eighth of a second and its mask follows it through the clip. The masks were drafted with SAM 2.1 from each object's tracked box and have not been corrected by hand.

Build the system, prepare the data, or both.

A Bengali delivery challan with seven fields labelled: vendor name, challan number, date, two line items, total amount and signature মেসার্স রহমান ট্রেডার্স ১৪ শাহজাদপুর, ঢাকা ১২১২ চালান নং ৪৮২১ তারিখ: ১২/০৯/২০২৬ বিবরণপরিমাণদরটাকা সুতি কাপড়১২০ মি.৩৫০৪২,০০০ বোতাম৪,৮০০ পিস১৪,৮০০ মোট ৳ ৪৬,৮০০ স্বাক্ষর vendor_name challan_no date line_item line_item total_amount signature
ai_engineering

AI engineering

We build AI features into your product and hand over code you own.

  • Document AI, including mixed Bengali and English
  • Computer vision: detection, segmentation, tracking
  • LLM assistants and agents connected to your tools
  • Voice AI for support calls and sales enquiries
  • Evaluation sets to judge the system against
AI engineering in detail
building
data_annotation

Data annotation

Our annotators label your data to your guidelines, and every batch is reviewed before delivery.

  • Images: boxes, polygons, segmentation
  • Video: objects tracked frame by frame
  • Text in English and Bengali
  • Documents: OCR correction and field extraction
Data annotation in detail Inspect our annotation example

Left: a fictional Bengali delivery challan with its fields labelled. Right: a still from the drone clip above, with one building outlined as a polygon.

Native Bengali, alongside your English-language work.

We tag Bengali and English text for classification, named entities and response rating, and we correct OCR on mixed-language documents. The sentence on the right is tagged the way a named-entity task would be.

ORGরহমান ট্রেডার্স DATE১২ সেপ্টেম্বর LOCঢাকার শাহজাদপুর থেকে QTY১২০ মিটার সুতি কাপড় পাঠিয়েছে।

"Rahman Traders sent 120 metres of cotton fabric from Shahjadpur, Dhaka, on 12 September." A fictional example.

Keep agreed work moving across time zones.

Dhaka is UTC+6, so US teams can hand over work in the evening and European teams share part of their day with us. We answer enquiries Monday to Friday in US Pacific working hours. Hand-off times for your project are agreed in its scope.

US PacificUTC-7
Central EuropeUTC+2
DhakaUTC+6
When we answer enquiriesLocal working day, for referenceNow, on Dhaka time

Start with a pilot. See the work as it develops.

  1. NDA

    We sign an NDA before we see any of your data.

  2. Paid pilot

    A separately scoped paid pilot comes before the main work.

  3. Batches and sprints

    Every annotation batch gets a QA review. Engineering work shows a demo every sprint.

  4. Your repository

    Code lives in your own repository from the start.

What are you trying to build or label?

Tell us the result you want, the kind of data involved and any deadline. Keep the first message general; we sign an NDA before you share data.

Discuss your project