Original Spec vs. Delivered Implementation
This matrix maps the original functional and technical requirements from the `GenAICodingExercise.pdf` against the actual shipped platform architecture.
Python REST API
Implement the solution in Python and create a REST API with endpoints to submit questions.
Decoupled Full-Stack Platform
Built the core intelligence using FastAPI in Python (/api/chat endpoint). Exceeded expectations by wrapping the API in a premium, fully-functional Next.js (React) Interface.
Dual Database Types
Use a Vector Database for unstructured data and a Relational Database for structured data.
Unified Supabase Infrastructure
Utilized Supabase to elegantly handle both within the same cluster. PostgreSQL strictly manages the relational equities, while the pgvector extension handles the 1536-dimensional math locally.
Hybrid AI Workflow
Combine insights from both data sources to provide comprehensive natural-language answers.
Context-Aware Orchestrator
Developed a robust custom Agent that calculates exactly when to deploy mathematical Vector Search and when to trigger safe Text-to-SQL logic against the tables, dynamically injecting a "Source Data Tag" in the UI.
No High-Level Frameworks
Do not use frameworks such as LangChain or LangGraph to solve the Orchestration.
Native Python Implementation
Created the entire pipeline "from scratch". Managed embedding extraction, custom chunking arrays, and Cosine Similarity algorithms manually with Python logic, proving deep architectural capability.
English Mandatory
All user interactions and responses should be in English.
Absolute Conformity
Configured the System Prompts and Next.js interfaces strictly to business-level English.