Proje: Okul Platform · Hub: Okul Platform — Decisions

TÜBİTAK KOBİ AR-GE #7260485 — Bidirectional AI Ecosystem Decision

Decision

Okul.com.tr will build a bidirectional self-learning AI ecosystem under TÜBİTAK KOBİ AR-GE Başlangıç Destek Programı (Project No: 7260485).

Project name: “Bidirectional Learning AI Ecosystem: Automatic Knowledge Loop with Smart Decision Support System from B2B-B2C Interactions in Education Marketplace”

Duration: 01.04.2026 – 30.06.2027 (15 months)

Budget: 3,496,170 TL (TÜBİTAK funded)

Company: Yeni Gelecek Bilişim Teknolojileri A.Ş. (operator of Okul.com.tr)

Why

The current platform runs on rule-based filtering + MySQL + Elasticsearch. No semantic search, no intent-based matching between parents and schools. B2B advisory is manual and unscalable. Knowledge gaps (questions users ask that the system cannot answer) are neither detected nor closed.

This project addresses three gaps:

  1. Semantic understanding via vector DB + embeddings for intent-based matching
  2. Dual-agent system: B2C parent counseling agent + B2B school coaching agent sharing a latent space
  3. Closed-loop knowledge system: every interaction generates new knowledge, the system self-updates

Team

RolePerson
Software Development SpecialistNuman Furkan Namlı (user)

Work Packages

WPScopeDuration
WP1Data Infrastructure, Knowledge Gap Detection & RAG Foundation01.04–30.06.2026 (90 days)
WP2Bidirectional GenAI Agent System Development01.07–31.10.2026 (122 days)
WP3Human-in-the-Loop Hybrid Model & Knowledge Loop01.11.2026–28.02.2027 (119 days)
WP4Behavioral Analytics & Predictive Insights01.03–31.05.2027 (91 days)
WP5Integration, Testing & Optimization01.04–30.06.2027 (90 days)

Success Criteria

  • Knowledge Gap F1 ≥ 0.75
  • Hallucination rate ≤ 5%
  • Cross-agent cosine similarity ≥ 0.85
  • HITL autonomous decision rate ≥ 80%
  • API response time < 2s

How to apply

This project adds a new service layer, new data infrastructure, and new AI components. Integration with the existing codebase will be via API adapters — no direct modifications to current source. See tubitak-arge-veri-altyapisi for the detailed data architecture plan.