Senior AI Engineer — Document Intelligence (Azure)
- Industry Other
- Category Software Engineering
- Location Kathmandu, Nepal
- Expiry date Jul 12, 2026 (1 day left)
Job Description
Heavy-lifts the build — extraction pipelines, validation agents, retrieval and integration — turning the architecture into working, tested, production-ready services.
Responsibilities
- Build and tune the extraction pipeline: custom Content Understanding analyzers / Document Intelligence models plus document-type classification and routing.
- Implement the clause-splitting and term-sheet correlation logic as agents / workflows (Foundry Agent Service, Agent Framework or LangGraph).
- Build retrieval and correlation over master data and rule libraries using Azure AI Search (vector + hybrid).
- Integrate ingestion (Power Automate / Logic Apps), expose APIs consumed by the Power Apps front end, and persist state in Dataverse.
- Implement confidence / grounding, discrepancy flagging and human-in-the-loop review hooks.
- Build the evaluation harness and observability / tracing; iterate prompts and models to meet accuracy targets.
- Apply secure engineering practices: managed identity, Key Vault, least-privilege access
Qualifications
- 5+ years software / AI engineering; strong Python (C# a plus).
- Production experience with Azure OpenAI / Microsoft Foundry and RAG using Azure AI Search.
- Hands-on document extraction with Azure AI Content Understanding and/or Document Intelligence.
- Building agentic / LLM workflows and integrating via REST APIs and SDKs.
- Prompt engineering plus practical evaluation of LLM output quality.
- Secure Azure development (managed identity, Key Vault, private endpoints).
Certifications
- Required or in progress: Azure AI Apps and Agents Developer Associate (AI-103). Existing Azure AI Engineer Associate (AI-102) holders welcome — the exam retired in June 2026, but the credential remains valid.
- Preferred: Microsoft Power Platform Developer Associate (PL-400).
Nice to have
- Power Platform / Dataverse integration; vector-search tuning; containerized deployment (AKS / Container Apps).
- Exposure to legal or finance document processing.
