What you'll build
A multi-agent AI layer that connects enterprise systems (NetSuite ERP, Oracle WMS, Base.vn CRM/Accounting, Microsoft 365, Google Workspace,…) and turns fragmented business data into actionable intelligence for CFO/CEO/COO — automated alerts, morning briefings, inventory insights, and cash flow forecasts.
1
Your role & responsibilities- Build and maintain LangGraph agents (ReAct pattern) that orchestrate multi-step reasoning across ERP, WMS, and accounting data.
- Develop MCP connectors (FastMCP 2.x) to integrate external business APIs — NetSuite SuiteQL, Base.vn, MS365, Google Workspace.
- Design agent skills (prompt engineering + output schemas) for inventory analysis, cash flow forecasting, dead stock detection, and supplier signal extraction.
- Build document ingestion pipelines: PDF, Word, Excel → structured data — layout parsing, table extraction, OCR for scanned documents.
- Build RAG pipelines to let agents retrieve from internal docs (SOPs, product catalog, historical alerts) to explain decisions and suggest next steps.
- Work with the LLM gateway (LiteLLM) to manage model routing, cost guardrails, and observability (Langfuse tracing).
- Build async Python services (FastAPI) and wire agent-to-agent communication via Redis Streams.
- Write unit tests for engine logic and integration tests with real Postgres + Redis (testcontainers).
2
Your skills & qualifications- 1–3 years Python experience, comfortable with async (async/await, httpx, FastAPI).
- Hands-on with LangGraph or LangChain agents — tool calling, state machines, multi-step reasoning.
- Understands LLM patterns: prompt engineering, tool use, structured output (Pydantic), context window management.
- Familiar with RAG: embeddings, vector search (pgvector / Qdrant), chunking strategies, retrieval quality evaluation.
- Experience with document processing: PDF/Word/Excel parsing, OCR (PaddleOCR / Surya), layout analysis (Docling, PyMuPDF), table extraction.
- Can read and integrate REST APIs with auth (OAuth1/2, API keys).
- Writes production-ready code: type hints, tests, clear error handling.
Nice to have
- Experience with FastMCP / MCP protocol.
- Familiarity with Redis Streams or other event buses.
- Exposure to multi-agent orchestration (supervisor pattern, A2A communication).
- Agent evaluation: LLM-as-judge, tool call precision, trajectory eval.
- Tool design patterns: idempotency, retry strategies, error recovery.
- Domain knowledge in ERP, inventory, or cash flow management.
3
Benefits- Compensation: Based on competence + Holiday/Tet bonuses linked to business performance
- Work alongside a team of experienced engineers and gain hands-on experience with real-world AIoT projects
- Full social, health, and unemployment insurance coverage in accordance with current regulations
- Clear career advancement path based on competence and performance
- Dynamic, proactive, and creative tech startup environment
- Opportunities to participate in industry events, seminars, training sessions, and team-building activities
