Self-Healing API Clients with Tiered Routing

POST /v2/orders
→ 422 Unprocessable
  "customer_id" unknown
  expected "customerId"

transform:
  customer_id → customerId
retry → 201 Created
cached for next call

API schema drift — renamed fields, changed payload shapes, new versions — breaks client requests at runtime and forces manual client-side patches.

Integrations fail silently or loudly until a developer investigates and patches the client. The research asks whether a client can detect, classify and adapt to these changes itself.

A middleware healing proxy sits between the client and the target API. It detects broken requests, classifies the failure and routes it by complexity: simple cases to a local model, complex cases to a cloud model. Repairs are grounded with a RAG pipeline over API specification knowledge and stored in a transform cache for reuse.

1. 01Runtime API / schema change
2. 02Failure detection
3. 03Failure classification
4. 04Tiered routing
5. 05LLM reasoning
6. 06Transformation / adaptation
7. 07Resilient API interaction

• Runtime schema/API change detection
• Failure classification
• Tiered routing: simple → Ollama (Llama 3 / CodeLlama)
• Tiered routing: complex → GPT-4o-mini
• RAG pipeline over specification knowledge (ChromaDB)
• Transform cache for repeated repairs
• Streamlit observability dashboard
• Automated evaluation harness

Individual M.Tech research in Software Engineering at BITS Pilani (WILP) — design, implementation and evaluation of the framework. This is an academic research prototype, not a production deployment.