redityaa/Qwen3-8b-BASE-SFT-V3
redityaa/Qwen3-8b-BASE-SFT-V3 is an 8 billion parameter Qwen3 model developed by redityaa, fine-tuned for automated verification and validation of eligibility for social assistance programs in East Java, Indonesia. It processes multi-role conversational messages to match citizen profiles against policy documents, specifically for the ASPD and PKH Plus programs. The model is designed to output pure JSON responses for audit and validation purposes.
Loading preview...
Model Overview
redityaa/Qwen3-8b-BASE-SFT-V3 is an 8 billion parameter Qwen3 model, fine-tuned by redityaa using Unsloth and Huggingface's TRL library. This model serves as an AI Auditor for the Social Affairs Office of East Java Province, designed to automatically, objectively, and data-drivenly verify and validate citizen eligibility for two specific social assistance programs: Asistensi Sosial Penyandang Disabilitas (ASPD) and PKH Plus (Lanjut Usia 70+).
Key Capabilities
- Automated Eligibility Verification: Processes multi-role conversational messages containing citizen profiles and policy context documents.
- Data-Driven Matching: Matches physical, economic, spatial, and functional parameters of citizen profiles against official policy rules.
- Strict Program Focus: Exclusively evaluates eligibility for ASPD and PKH Plus, adhering to strict input guardrails that prohibit recommendations for other programs.
- Structured JSON Output: Generates pure JSON object responses without markdown blocks, providing a
ringkasan_profil(profile summary),rekomendasi(eligible/possibly eligible programs),rekomendasi_teknis_bansos(operational action plan), andprogram_tidak_sesuai(ineligible programs with reasons).
Use Cases
This model is specifically engineered for government agencies or organizations requiring automated, objective auditing of social welfare program eligibility based on predefined policy documents and citizen data. Its strict JSON output format makes it suitable for integration into automated decision-making or reporting systems.