LangChain is not the workflow engine. In ClaimFlow AI, LangChain is used to connect:
claim context + model + tools + structured output
LangChain helps with:
model calling
tool definitions
structured action schemas
agent-style planning
controlled tool orchestration
LangChain should not replace:
Prisma database
review queue
validation rules
RAG citation guardrails
business workflow state
human review decisions
The model is the reasoning layer.
In ClaimFlow AI, the model answers:
Given this claim state, what should happen next?
Examples:
Missing FIR number → request missing evidence
Weak policy evidence → escalate to human
Repair estimate alone → mark needs more evidence
Review already pending → no action
The model should suggest actions, not directly approve or reject claims.
Context is the claim state passed to the model.
In ClaimFlow AI, context comes from buildAgentContext(runId).
It includes: