The project and its needs
FaRkLaR.net is a guide to the differences in meaning between Turkish words and concepts. Its core unit is not a single word but two or more terms and the difference between them. Compiled by hand since 1996, it now holds 207,622 difference entries, 137,283 explanations and 112 language definitions.
At this scale the content had to be fast, searchable and browsable through its relations for readers, and consistent and manageable for editors.
Data model: relation operators and hierarchy
Entries are built with typed relation operators: parallel (||), equal (=), opposite (<>), direction (>), "instead of", "not", etymology and same spelling. First, middle and last letter or word variants can also be searched as separate relation types.
- A tree of 306 guides (26 main + 280 sub) and 2,222 topics
- One entry can belong to several guides
- 112 languages defined, equivalents recorded in 94
- Side modules for units of measure, basic sciences and settlements
Reader features
- Quick, advanced and full-text search
- Autocomplete, instant search and voice search
- An interactive term graph and visual map on every entry
- Difference, quote, story, film and book of the day; daily archive and RSS
- Membership, Google sign-in, ratings, comments and favourites
- Installable app (PWA) with an offline page
Editing and administration
The admin panel has about 55 pages. Entries added by users and suggestions imported from other dictionaries go through an approval queue; editors add entries in bulk with a compact line-prefix markup.
- A "Fill" button that completes missing fields from similar entries
- Automatic validation and a Turkish style corrector
- Duplicate-cluster finder with bulk delete
- Share-card image generator, statistics dashboards and backup manager
AI, MCP and API
When a reader asks a question, the answer is generated from entries retrieved from the database and comes with numbered sources (RAG). Gemini is available to the public; Claude, ChatGPT, Grok and DeepSeek are available to members.
- Remote MCP server with search, entry lookup, word comparison and guide index tools
- OAuth 2.1 + PKCE and daily quotas by membership tier
- REST API with an OpenAPI specification
- llms.txt and llms-full.txt; a separate licence for AI training
Search visibility and infrastructure
Its 213,844 URLs are regenerated daily in sharded sitemaps, five of them with 50,000 URLs each. Caching, bot protection and an allowance for AI crawlers verified by their official IP ranges carry heavy traffic.
What this means for your project
The relation model, editor panel, RAG answers and MCP server proven on FaRkLaR.net can form the base of your dictionary too, in less time and at lower cost than building from scratch.