A ANKA DATATurkish AI Data & Outsourcing
Turkish AI Data Collection & Outsourcing

Human-generated Turkish data for the AI models of tomorrow.

ANKA DATA is a dedicated team based in Turkey that produces high-quality Turkish-language data for AI training and evaluation. You tell us what you need — we build the team, run the project, and deliver. Whether it is a custom data collection task or a team of Turkish speakers working directly on your platform, we make your AI better in Turkish.

Start a project See what we can do
Audio · Video · Image · TextEvery modality, one Turkish-speaking team
Task-driven deliveryYour spec, your format, your quality bar
Two ways to workCustom data tasks or embedded on your platform
Based in TurkeyNative speakers, reliable and cost-effective
How we work

Two ways to partner with ANKA DATA

AI companies and data vendors work with us in the way that fits their pipeline. Many clients use both models at the same time.

Model 1

Custom Turkish data collection tasks

You send us a task, and we deliver the data. Our team plans, recruits, records, reviews and ships exactly what the brief asks for — no generic, purpose-less datasets. We start from your requirements: formats, quality standards, volumes, and deadlines.

  • “Record 3 hours of natural conversation between 2 people.” — done.
  • “Collect 50 speakers reading a fixed prompt list in different accents.” — done.
  • “Film 20 people doing everyday activities.” — done.
  • “Photograph 500 different objects from multiple angles.” — done.
  • If it is a Turkish-language data task, we can scope it, staff it and deliver it.
Model 2

Turkish AI model improvement on your platform

Our trained Turkish-speaking team works directly inside your tools and platforms to improve your AI models’ Turkish performance: following your guidelines, using your workflows, and reporting to your project managers.

  • Rating, ranking and comparing Turkish model responses
  • Writing high-quality Turkish prompts and ideal responses
  • Finding errors, hallucinations, and unnatural phrasing
  • Safety, red-teaming and cultural-nuance review in Turkish
  • Annotation, labeling and quality review at scale
Model 1 · Custom data collection

Which tasks can clients give us?

If a client can describe a Turkish-language data need, we can organize a team to produce it. Below is a detailed look at the task types we take on — and these are only examples. Custom and hybrid tasks are welcome.

🎙️ Audio & Speech Tasks

Human voice data in Turkish for speech recognition (ASR), text-to-speech (TTS), voice assistants, and conversational AI.

  • Natural conversation recordings — e.g. two people talking naturally for 3 hours; unscripted, spontaneous, on topics you choose (or free topics).
  • Scripted reading — e.g. 50 speakers reading a fixed sentence list, to capture the same content across many voices.
  • Accent & dialect coverage — speakers from different regions of Turkey, ages and genders, with speaker metadata.
  • Multi-speaker dialogues — discussions, debates, interviews, podcast-style chats, and meeting-style conversations with 3+ participants.
  • Emotional & expressive speech — excited, calm, angry, sad, sarcastic or neutral delivery, recorded to a defined emotion spec.
  • Phone-call simulations — customer service, appointment booking, sales and support calls with role-play scenarios.
  • Voice commands & wake words — smart-home, in-car and assistant commands, spoken in many natural variations.
  • Environment & device diversity — quiet room, street, café, car, far-field and different phones or microphones.
  • Studio-quality voice recording — clean, consistent recordings for TTS voice building (with speaker consent).

🎥 Video Tasks

Real-world Turkish video for computer vision, action recognition, egocentric AI, robotics and video-language models.

  • Daily-activity videos — e.g. 20 people recording their everyday routines, first-person or third-person.
  • Cooking & food preparation — Turkish and international recipes filmed step by step.
  • Object usage demonstrations — how tools, appliances and products are used in real life.
  • Human motion & hand gestures — hand–object interaction, gestures, body movement, and posture.
  • Multi-angle object capture — the same object or scene filmed from specified angles, distances and lighting.
  • Talking-head & expression videos — people speaking to camera for audio-visual and lip-sync research.
  • Instructional & how-to videos — tasks performed with spoken explanations in Turkish.
  • Indoor & outdoor scenes — homes, workplaces, shops and streets, with privacy and consent rules respected.
  • Turkish Sign Language (TİD) — scoped per project, depending on signer availability.

📷 Image Tasks

Photographs and labeled images for classification, detection, OCR and image–text models.

  • Object photography — e.g. 500 different objects photographed to a defined shot list.
  • Product capture — specific products from multiple angles, backgrounds and lighting conditions.
  • Environment & scene photos — kitchens, living rooms, offices, shops, streets and more.
  • Local-context imagery — Turkish signage, packaging, food, storefronts and everyday items that global datasets often miss.
  • Classification & tagging — sorting images into categories and adding attributes or tags.
  • Bounding boxes & segmentation — marking objects and regions with precise annotations.
  • Image captioning in Turkish — short and detailed natural-language descriptions written by native speakers.
  • Turkish OCR & handwriting — handwritten samples, printed text photos, and text-in-image transcription (including special characters ç, ğ, ı, ö, ş, ü).
  • Screenshot & UI data — Turkish app and website screens with annotation.

✍️ Text & Language Tasks

Written Turkish created, edited and judged by real people — for LLM training, fine-tuning and evaluation.

  • Question–answer creation — original Turkish questions with accurate, well-written answers.
  • Prompt & instruction writing — diverse, realistic user prompts for instruction tuning.
  • Free writing on given topics — articles, essays, stories, emails, reviews, and social-style posts.
  • Multi-turn chat dialogues — realistic Turkish conversations between a user and an assistant.
  • AI answer evaluation — rating, ranking and comparing model responses against guidelines.
  • Grammar, spelling & style correction — fixing errors, improving fluency and keeping the right register.
  • Translation & localization — English ⇄ Turkish translation and cultural adaptation, not just word-for-word.
  • Summarization & rewriting — shortening, paraphrasing and tone changes (formal, casual, slang).
  • Classification & labeling — sentiment, intent, topic, toxicity and named-entity annotation.
  • Fact-checking — verifying claims in Turkish text with sources.
  • Reasoning & math prompts — problems and step-by-step solutions written and checked in Turkish.
  • Domain-specific writing — education, finance, law, health, business and more, depending on specialist availability.

🧩 Multimodal Tasks

Combined data types that teach models to connect what they see, hear and read.

  • Activity + audio + description — a video of a person doing something, with their natural voice and a written explanation.
  • Speech paired with transcripts — recorded audio aligned with accurate Turkish text.
  • Image/video + spoken description — people describing what they see out loud.
  • Voice questions about visuals — a person asks a spoken question about an image or video and answers it.
  • Video object & action labeling — tagging objects, hand movements and events over time.
  • Screen recordings with narration — step-by-step computer or phone use explained in Turkish.
  • Custom multimodal datasets — any combination of audio, video, image and text designed to your schema.

🏷️ Transcription & Annotation Tasks

Turning raw data into structured, model-ready training data.

  • Turkish transcription — verbatim or clean-verbatim, with timestamps and your style guide.
  • Speaker labeling (diarization) — who spoke when in multi-speaker audio.
  • Audio event & noise tagging — laughter, background noise, overlaps, music and other events.
  • Emotion & intent labeling — tagging how something is said and why.
  • Speech quality review — scoring recordings for clarity, naturalness and pronunciation.
  • Data cleaning & validation — reviewing existing Turkish datasets, removing errors, duplicates and low-quality items.
Pronunciation datasets Persona / role-play dialogues Slang & informal Turkish Call-center scenarios Reading comprehension sets Document & form samples Synthetic data review Benchmark creation Custom pilots & samples
Model 2 · Work on your platform

A Turkish-speaking team that plugs into your workflow

Instead of delivering files, our team works inside your annotation tools, projects and platforms — improving your AI model’s Turkish performance directly. You keep your processes; we supply trained, reliable people and a team lead who manages quality and communication.

⚖️

Response evaluation & ranking

Compare two or more model answers, score them on your rubric (accuracy, helpfulness, fluency, tone) and justify the decision in writing.

🧠

RLHF & preference data

Produce human feedback data that teaches models what a good Turkish answer looks like — including ideal “gold” responses written from scratch.

🇹🇷

Linguistic & cultural review

Catch unnatural phrasing, wrong register, mistranslations, grammar and agglutination mistakes, and culturally inappropriate or inaccurate content.

🛡️

Safety & red-teaming

Test models with adversarial Turkish prompts, flag unsafe or biased outputs and document failure cases following your safety policy.

🔍

Fact-checking & hallucination detection

Verify claims in model responses against reliable sources and mark errors with clear explanations.

✍️

Prompt & response writing

Create diverse, realistic Turkish prompts and expert-quality responses for supervised fine-tuning.

🎧

Voice & audio model evaluation

Judge speech recognition, TTS naturalness and voice-assistant behavior in Turkish, including accents and noisy conditions.

🖼️

Multimodal evaluation

Review how well models describe, answer questions about, and reason over images, video and audio in Turkish.

🎓

Domain expert review

Subject-matter reviewers for fields like education, law, finance, health and business — scoped per project and availability.

Process

From first message to delivery

A simple, transparent process built around your requirements.

Brief

You share your needs: data type, formats, quality standards, volumes, timeline and budget. We ask clarifying questions before anything starts.

Pilot / sample

We produce a small batch (or onboard a few team members) so you can check quality and adjust the spec before scaling.

Team setup & training

We recruit the right participants or annotators in Turkey, train them on your guidelines and set up consent and onboarding.

Production

The team starts collecting or working on your platform, with a project lead tracking progress daily.

Quality review

Multi-stage checks, spot audits, and rework of anything that misses the bar — before it reaches you.

Delivery & iteration

Data is delivered in your required format with metadata and a summary report. We use your feedback to improve the next batch.

Quality, formats & trust

Built for clients who care about data quality

Human-generated data is only valuable if it is consistent, well-documented and responsibly collected.

Quality control

  • Guideline training before production
  • Reviewer layer on every batch
  • Spot checks and per-contributor scoring
  • Rework of rejected items at no extra effort for you

Flexible deliverables

  • Audio: WAV / FLAC / MP3, sample rate and bit depth to spec
  • Video: MP4 / MOV, resolution and frame rate to spec
  • Images: JPG / PNG / RAW as required
  • Text & metadata: JSON, JSONL, CSV, XLSX

Consent & privacy

  • Informed consent from every contributor
  • Speaker and device metadata only as agreed
  • Confidential handling and NDA available
  • Collection practices designed with KVKK and GDPR principles in mind

Why Turkish data matters

Turkish is spoken by tens of millions of people and has rich dialect variety, a complex agglutinative grammar and strong cultural context. Global models often underperform in it because high-quality, human-generated Turkish data is limited. Native speakers working under clear guidelines close that gap.

Why ANKA DATA

We are a focused Turkey-based team: native speakers, diverse regions and demographics, a time zone that overlaps with both Europe and the Americas, and a cost-effective model. We prefer to understand your exact data needs first, then build the team and workflow around them.

FAQ

Frequently asked questions

Who do you work with?

AI labs, AI data and annotation companies, research teams and technology businesses that need Turkish-language data or Turkish-speaking contributors — as a direct supplier or as a subcontracting partner.

Do you sell ready-made datasets?

Our focus is purpose-driven production: we collect and create data against your specification. We prefer to understand your requirements first rather than produce generic datasets without a defined use.

Can you work in our tools and platforms?

Yes. Our second model is exactly that: a trained Turkish team working inside your annotation tools and projects, following your guidelines and reporting to your project managers.

What if our task isn’t listed here?

The lists above are examples. If it involves Turkish language, speech, video, images or text produced or reviewed by humans, send us the brief and we will tell you honestly what is feasible, how, and at what cost.

How do you handle pricing, volumes and deadlines?

We discuss formats, quality standards, volumes and timelines with you, then quote per task, per hour of data, per unit or per team-hour depending on what suits the project. A pilot batch is a good way to start.

How is contributor privacy handled?

Every contributor gives informed consent, and we collect only the metadata your project requires. We can sign an NDA and follow your data-handling requirements.

How do I start?

Message or email us with a short description of your need. We reply with questions, a plan and, if useful, a pilot proposal.

Let’s work together

Have a Turkish data task or a model to improve?

Tell us what you need — data type, volume, formats and timeline — and we will come back with a clear plan. You can reach us on WhatsApp or by email.

WhatsApp

+1 227 264 2886 Chat on WhatsApp

Email

ulualiy@zohomail.eu Send an email