Valura AI Engineer Assignment - AI Co-Investor Microservice Defense | Nihal Jaiswal
This is my technical defense video for the Valura AI Engineer (Round 2) project assignment.
Project Overview:
I have engineered a production-ready "Microservice Spine" for an AI Co-Investor platform. The system is designed to be extensible, cost-effective, and highly secure, featuring a multi-stage pipeline that handles everything from financial compliance to real-time data streaming.
Key Features & Performance:
Safety First: A deterministic, regex-based Safety Guard covering 7 financial risk categories with 100% recall.
High-Accuracy Routing: Intent Classifier achieving 98.4% accuracy using OpenAI structured outputs.
Real-Time UX: Full implementation of Server-Sent Events (SSE) for streaming classification and results.
Implemented Agents: Fully functional Portfolio Health agent (via yfinance) and a deterministic Financial Calculator.
Testing Rigor: 56/56 tests passing with a mocked LLM environment for CI/CD reliability.
Video Timestamps:
Introduction & Project Scope
Architecture Deep Dive (The Journey of a Request)
Live Demo: Pytest Suite (56 Tests Passing)
Live Demo: API Endpoint & SSE Streaming
Safety Guard: Blocking Harmful Financial Intent
Performance Metrics (Accuracy, Latency, & Cost)
Non-Obvious Decision: Why Regex over ML for Safety?
Future Scalability: Redis Sessions & Embedding Pre-Classifiers
Final Summary & Closing
Tech Stack: Python, FastAPI, OpenAI API (GPT-4o-mini), Pydantic, yfinance, Pytest, SSE.