Technical Specification

TECHNICAL SPECIFICATION

High-Performance Implementation for AI Capital Research

1. Architecture Overview

Presentation

CLI (Python) / Web (Next.js)

Orchestration

FastAPI / Custom Python Logic

Data Access

Supabase / pgvector

Data Source

PostgreSQL (1,821 stocks)

// System Sequence Flow

[1] User Input "Amazon target price + inflation risks"

[2] Classification LLM detects "Hybrid" intent

[3] Entity Extract "AMZN" | "Amazon.com, Inc."

[4] SQL Search Parallel query to equities table

[5] Vector Search Embedding generation + Similarity search

[6] Synthesis GPT-4o-mini combines structured + unstructured data

[7] Return Answer + Source: Hybrid

2. Core Components

Relational DB

Supabase (PostgreSQL) optimized for financial lookups.

CREATE TABLE equities (
  ticker VARCHAR(10) PRIMARY KEY,
  name TEXT,
  target_price NUMERIC,
  market_cap NUMERIC,
  dividend_yield NUMERIC,
  sector_level1 TEXT
);

Vector Database

pgvector extension with 1536-dim embeddings.

CREATE OR REPLACE FUNCTION match_documents(...)
RETURNS TABLE(...)
ORDER BY embedding <=> query_embedding
LIMIT match_count;

Models & LLM Integration

Entity & Classification

GPT-4o mini

TEMP: 0.0

Vectorization

text-embedding-3-small

TEMP: N/A

Final Synthesis

GPT-4o mini

TEMP: 0.1

Rigor. Accuracy. Traceability. GenAI Capital v1.0