Xplora Kodo — Agentic AI Platform for Workforce Migration

Problem

Migrating workers from Nepal into the Japanese workforce involves a genuinely hard set of interlocking requirements: visa compliance that varies by category, language proficiency that has to reach a working standard, interview readiness, financial tracking across the process, and job placement — each with its own domain expertise. A single monolithic chatbot can't hold that much specialized knowledge coherently, and a team of disconnected point tools doesn't give candidates a single guided journey.

Action

• Designed and built a multi-agent swarm using Agency-Swarm, with 17+ specialist agents each owning a distinct part of the journey: visa compliance checking, interview proctoring, career placement matching, financial ledger tracking, and RAG-backed language coaching.
• Built the backend on FastAPI with PostgreSQL + pgvector, giving the RAG-based language coaching agent fast semantic retrieval over training material and past interactions.
• Implemented WebXR-based AR/VR training modules with gaze-triggered questioning — letting candidates practice interview scenarios in an immersive environment where the system tracks what they're looking at to time its prompts.
• Coordinated agent hand-offs so a candidate's journey stays coherent even as different specialist agents take over different stages — visa status feeding into interview readiness, interview performance feeding into placement matching.

Result

A single platform that replaces what would otherwise be a fragmented set of manual processes and disconnected tools with one coordinated, AI-guided journey from language readiness through visa compliance to job placement — built on an architecture (specialist agents over a shared knowledge base) that can extend to new destination countries or worker categories without a rewrite.