AskAra+

Migrant workers moving between ASEAN countries run into the same wall over and over: government service portals that assume you already speak the local language and already know the bureaucracy. AskAra+ removes that wall.

The problem

Work permits, healthcare enrolment, and legal aid are all gated behind government documentation that’s rarely translated, rarely explained, and rarely designed for someone arriving with no local context. Existing chatbots either hallucinate procedure or can’t hold a conversation in the right language.

What we built

Built together with my co-founder Lineysha, who led the AI/ML side while I led the full-stack build. She grounded every answer in a ChromaDB retrieval layer built over 2,286 real government documents on top of SEA-LION v4 — a language model trained specifically on Southeast Asian languages — pairing each document with a per-country jargon glossary so a term like “e-KTP” or “EPF” resolves correctly instead of being mistranslated.

On the full-stack side, I built a deterministic agent pipeline — not a free-roaming LLM chat loop — orchestrating an 11-tool FastMCP toolchain around that retrieval layer, plus OCR for scanning physical documents on the spot and voice input/output via Groq Whisper for users who’d rather speak than type. Together, we also built an eligibility profiler that narrows down which services actually apply to a given worker’s situation. The whole thing ships as an installable PWA so it works like a native app on a borrowed or low-spec phone.

Outcome

AskAra+ reached the finals of V Hack 2026 at Universiti Sains Malaysia, built and shipped with Lineysha — the RAG pipeline and SEA-LION integration on her side, the FastMCP tool orchestration and deployed frontend on mine.