新北市汐止區5年以上大學以上
【Role Overview】
Be part of a Pixel-team POC that ingests client device data, powers a multi-agent AI assistant using Google ADK/Vertex AI with RAG, and surfaces compliance insights via a simple web UI. Own data pipelines, agent orchestration (MCP), LLM integration, and lightweight dashboard/reporting.
【Key Responsibilities】
Data & ETL:Ingest CSV/Excel/Docs/PDF from GCS; normalize, chunk, embed; incremental sync to vector store.
Agentic AI:Design MCP hub-and-spoke; build specialized agents (RAG-Fusion KB, CertCloud, RF-OTA reporting, test automation, lab scheduling).
RAG & LLM:Implement retrieval chains, prompt design, context assembly, quality/latency tuning.
Web UI:Deliver minimal Django UI for file upload + chat.
【Must-Have Qualifications】
- 5+ years of professional full-stack development experience, followed by 1+ years of hands-on AI agent development.
- Proven track record of shipping production AI agents with at least a few hundred active users.
- Direct contributor to the core agent logic, not just peripheral tooling or integration.
- Strong Python skills.
- Experience with agent frameworks and building AI Agents (Google ADK, LangChain, LangGraph, LlamaIndex, Semantic Kernel).
- Hands-on AWS/Azure/GCP (preferred) such as Cloud Storage, Pub/Sub, Cloud Functions/Run, IAM, and Vertex AI.
- RAG experience end-to-end (chunking, embeddings, vector search, fusion/re-ranking, prompts).
- Understanding of data quality impacts on AI responses, including hallucinations, context loss, and deployment best practices.
- Vector search on GCP (Matching Engine or BigQuery Vector Search, or equivalent ANN experience).
- Frontend familiarity to wire a minimal chat/upload UI and embed dashboards.
- Bilingual Mandarin/English; comfortable collaborating with Banqiao and US teams.
【AI Agent Experience Requirements】
- Demonstrated ability to design, deploy, and scale agentic AI systems in production.
- Prefer candidates who have implemented MCP (Model Context Protocol) and/or multi-agent architectures.
- Clear understanding of how data pipelines, embeddings, and retrieval strategies affect agent performance and user trust.
【Nice-to-Have】
- Scheduling/optimization know-how.
- Telecom/test reporting exposure (RF/OTA) or QA automation integration.
- Security/compliance on GCP (VPC-SC, org policies).