Own product · Legal AI
Ferman: legal research assistant for Turkish law
Ferman searches Turkish legislation, case law and doctrine and gives answers in which every sentence points to a source. If it can't find something, it says so — and a fabricated statute never gets through.
- 480,000+legal documents
- 4M+searchable text passages
- 1M+citation links between sources
- 10 / 10correct refusals, 0 fabricated citations in the refusal test
Status September 2026
Why Ferman
Legal research in Turkish law means checking statutes and amendments, searching tax rulings, comparing Danıştay and Yargıtay case law and following chains of references. General AI chatbots speed this up, but sometimes invent articles or cite laws that were repealed years ago. Ferman is built to prevent exactly that.
What Ferman does
Answers with citations
Every answer points to the exact article, decision or ruling, with a link to the source text.
Checks for fabricated citations
Every citation in an answer is checked against the database. Anything that doesn't exist is flagged instead of presented.
Knows when something doesn't exist
Ask for an article above a law's highest article number and Ferman says that article doesn't exist — and names the highest number.
Recognises repealed provisions
When a repealed article is cited, Ferman reports the repeal date and its successor, and bases the answer on the rule in force.
Versions over time
Statutory articles are stored with validity dates, so you can ask which text applied on a given date.
Research notes
Results can be exported as a research note (.docx) with a source list; drafts are always marked as drafts.
Sources in the database
Legislation
Laws, regulations and communiqués from the official Turkish legislation services, including repealed laws such as the old HUMK, TTK and TKM
Tax rulings
Rulings (özelge) from the Turkish tax administration GİB
Supreme courts
Decisions of the Danıştay and Yargıtay, including unifying judgments (İBK)
Courts of appeal
Decisions of the Bölge Adliye Mahkemeleri (istinaf)
Constitutional Court
17,067 individual applications and 5,606 norm reviews of the AYM
European Court (ECtHR)
6,670 Turkish-language documents from HUDOC plus the official Turkish text of the ECHR and its protocols
Competition
Law 4054, secondary legislation and Rekabet Kurulu decisions
Doctrine
Legal handbooks and articles
How an answer is produced
Privacy first
Personal data such as names, ID and IBAN numbers are masked before anything goes to a language model.
Hybrid search
Semantic (vector) search, Turkish full-text and the citation network between sources in a single query.
Reranking
A reranker picks the most relevant passages, giving statutes priority over decisions.
Answer from sources
The model answers only on the basis of the retrieved passages.
Citation check
Every citation is verified; repealed or non-existent articles are reported.
Recorded
Every query goes into an audit trail that cannot be altered afterwards.
Measured, not guessed
Each area of law has its own test set of questions whose correct source was verified in advance. Every change is measured on hit@1, recall@5 and MRR, and a separate refusal test checks that Ferman refuses — instead of inventing — when the sources hold no answer.
Security & hosting
- Hosted in the EU (Hetzner, Germany)
- No public access: private network only (Tailscale)
- Personal data masked before language-model processing
- Audit trail that cannot be altered afterwards
- Infrastructure as code (Terraform)
Important
Ferman is a research tool, not a lawyer. It gives no legal advice and makes no decisions; the final word always rests with a legal professional who checks the sources.
Turkish law: legal support & AINeed a legal AI assistant of your own?
We build similar research assistants on your own documents and sources — in the EU or fully local.