Internship - Conversational AI - Context Aware Assistant (H/F)

iWE is an Advanced Claim Management platform. Case folders aggregate structured data, documents, and workflow state that users need to understand quickly. The purpose of this internship is to design and prototype a conversational assistant embedded in the case management application that leverages case-specific context (metadata, related entities, and relevant content) so users can ask natural-language questions and receive accurate, grounded answers and reduce time spent searching.

The scope also covers integration constraints with the existing iWE architecture (security, permissions, and data boundaries), evaluation of LLM and retrieval approaches (MCP, tooling, guardrails), and a working demonstration inside a representative case experience.

This is a 3-month internship for summer 2026.

What You’ll Do

• Work with a Product Owner and engineering team to clarify use cases (e.g. “What is the current status?”, “Summarize recent activity”, “Explain this field”) and non-functional requirements (latency, auditability, privacy).

• Map what “context” means in iWE: which objects, fields, events, and documents can feed the assistant, and how access control must be enforced per user and per case definition.

• Research and compare implementation options: LLMs, MCP, retrieval-augmented generation, function calling against APIs, and evaluation metrics.

• Design the assistant’s interaction model (entry point in the case UI, conversation history, citations to source records where applicable).

• Implement a proof-of-concept: ingest or query case context safely, connect to a conversational layer, and expose answers through a minimal integration path aligned with iWE’s stack and patterns.

• Document limitations, risks (hallucinations, PII), and a roadmap for hardening (logging, feedback, admin controls).

Technologies / Challenges

• Large language models and prompt / tool orchestration.

• Retrieval and grounding (semantic search, embeddings, chunking, citation).

• REST or event-driven APIs; possible WebSocket for streaming replies.

• Security: authentication, authorization, tenant and folder isolation.

• UX for conversational UI in a business-critical claims context.

• Performance and cost trade-offs (token limits, caching, summarization).

Desired Skills and Experience

• Solid programming fundamentals; experience with JavaScript/TypeScript or similar backend stacks is a plus.

• Interest in NLP / LLM applications and willingness to read papers and product docs.

• Understanding of web APIs, data modeling, and basic security concepts (least privilege, secrets).

• Curiosity about low-code / enterprise software and regulated domains (insurance / claims).

• Autonomous, structured, and passionate about building reliable prototypes.