Project Case Study
Semantic API
Discovery Platform
A semantic search system that lets developers find API documentation by describing what they need — built for a major Malaysian financial institution.
Case Study
Problem & Solution
The Problem
Developers navigating a large financial institution's API catalogue had to scroll through pages of documentation or rely on exact keyword matches to find what they needed. This created friction, slowed development velocity, and led to APIs being underutilised simply because they were hard to discover.
The Solution
A semantic search system that understands developer intent. Ask “I want to find Authentication APIs” and the system returns a ranked recommendation with alternatives and a plain-English justification explaining why each result relates to the query — no exact keywords needed.
Architecture
System Diagram
← scroll to pan →
How It Works
The Search Pipeline
MongoDB
API Data Source
OpenAI Embeddings
text-embedding-3-large
ChromaDB
Vector Store
HuggingFace Reranker
25 → top 5 results
GPT-4o
Response Generation
Redis Cache
10-min TTL
MongoDB
API Data Source
OpenAI Embeddings
text-embedding-3-large
ChromaDB
Vector Store
HuggingFace Reranker
25 → top 5 results
GPT-4o
Response Generation
Redis Cache
10-min TTL
Capabilities
Key Features
Semantic Intent Search
Find APIs by describing what you need in plain English, not exact keywords. The system vectorizes the query and performs similarity search across the full API catalogue.
Contextual Recommendations
Returns a main recommendation plus alternative options, each with a plain-English justification explaining exactly why it matches the query.
HuggingFace Reranking
Vector search retrieves 25 candidates; a HuggingFace reranking model reranks them to the top 5 most relevant results before passing to the LLM.
Live Data Sync
MongoDB replica-set with change-stream enabled automatically keeps ChromaDB in sync as API documentation is updated — no manual re-indexing.
Redis Response Caching
Identical queries are served from cache for 10 minutes, reducing LLM calls and improving response latency for repeated searches.
GPT-4o Grounded Generation
Responses are grounded strictly in retrieved API data. The LLM formats a tailored answer with ranked suggestions and rationale, not hallucinated content.
Stack
Technology
API Layer
Database
Embedding
Vector Store
Reranking
LLM & Cache