Viewing context:Principal Architect

AI-Architected Project

"100% AI-Orchestrated with human direction"

AI Agent: Resume

Elegant Software Solutions

Intelligent AI assistant for recruiter engagement, powered by Claude 4.5 Opus. Features sophisticated RAG-based semantic search, intelligent JD analysis with role detection, persistent cross-session memory, and multi-role resume adaptation. Enables natural conversations about professional background while learning recruiter preferences.

753

JD Analyzer LOC

343

RAG Search LOC

100+

Tech Patterns

Role Perspectives

RAG & AI System Design

Contribution as Principal Architect

Designed hybrid RAG architecture with vector + keyword retrieval
Created JD analysis engine with 100+ technology patterns across 6 categories
Architected two-tier fact verification with static and dynamic layers
Designed recruiter memory system with HMAC-secured sessions

Architecture Overview

AI Resume Agent Architecture:

┌─────────────────────────────────────────────────────────────┐
│  FRONTEND: Floating Chat (React + Framer Motion)            │
│  SparklesChatContext → FloatingAIChat → Streaming Display   │
└─────────────────────────────────────────────────────────────┘
                          ↓ HTTP/SSE
┌─────────────────────────────────────────────────────────────┐
│  API: /api/resume-chat (Next.js API Route)                  │
│  Session Management → Context Building → Claude 4.5 Opus    │
└─────────────────────────────────────────────────────────────┘
                          ↓
┌─────────────────────────────────────────────────────────────┐
│  JD ANALYSIS: Intelligent Processing (753 LOC)              │
│  Technology Detection → Role Scoring → Match Analysis       │
└─────────────────────────────────────────────────────────────┘
                          ↓
┌─────────────────────────────────────────────────────────────┐
│  RAG LAYER: Hybrid Search (343 LOC)                         │
│  Vector (pgvector) + Keyword (ILIKE) + Facts (TypeScript)   │
└─────────────────────────────────────────────────────────────┘
                          ↓
┌─────────────────────────────────────────────────────────────┐
│  DATA: Supabase                                             │
│  resume_embeddings | resume_facts | memory_sessions         │
└─────────────────────────────────────────────────────────────┘

Core Capabilities

Conversational Resume Q&A

Natural language understanding of recruiter questions with proactive surfacing of relevant achievements and intelligent handling of follow-ups.

JD Match Analysis

Automatic analysis of uploaded job descriptions with technology detection, requirement extraction, role recommendation, and match scoring (Strong/Solid/Moderate/Stretch).

Hybrid RAG Search

Combined vector embeddings + keyword search across categories: master-documents, interview-prep, technical deep-dives, project case studies.

Persistent Memory

Remembers recruiters across visits without login—stores preferences, chat history, notes, and automatically extracts recruiter preferences from conversation.

Role-Specific Framing

Adapts response perspective based on selected role—executive strategy for CTO, hands-on implementation for Developer, team leadership for Director.

Multi-Format JD Support

Accepts PDF, DOCX, Markdown, and plain text job descriptions for analysis with intelligent parsing and technology extraction.

Real-Time Streaming

Token-by-token response streaming for natural conversation flow with animated UI feedback.

Lead Capture & Notification

Integrated contact collection with instant Slack alerts and email notifications when recruiters express interest.

Modules & Services

Claude 4.5 Opus Integration

Powered by Anthropic's most capable model (claude-opus-4-5-20251101) for nuanced, context-aware conversations with role-specific response framing and intelligent follow-up handling.

RAG Pipeline

Retrieval-Augmented Generation with OpenAI text-embedding-3-small (1536 dimensions), pgvector similarity search, and configurable threshold retrieval. Top 8 results with category filtering.

JD Analysis Engine

Intelligent job description processing (753 LOC) with technology detection (100+ patterns across 6 categories), requirement extraction, seniority detection, and automatic role classification with match scoring.

Recruiter Memory System

Persistent session memory with 90-day cookie-based identification, preference extraction (remote, location, company size, industry), chat history (last 20 messages), and personal notes storage.

Fact Verification System

Two-tier fact storage: static TypeScript layer for core facts and dynamic Supabase layer for updatable content. Cross-references all responses against verified facts database.

Multi-Role Adaptation

Dynamically adapts responses based on target role perspective (CTO, VP, Director, Architect, Developer). Same career history, five professional viewpoints tailored to audience.

Streaming Chat Interface

Real-time token streaming with Framer Motion animations, floating chat UI, and Server-Sent Events for natural conversation flow.

Lead Capture Integration

Contact collection modal with multi-channel notification to Slack and email. Captures recruiter details, company info, and optional JD uploads.

Tech Stack

Languages

TypeScriptSQL

Frameworks

Next.js 16React 19Vercel AI SDK

Infrastructure

VercelSupabase CloudOpenAI APIAnthropic API

Data

SupabasePostgreSQLpgvector

Loading project...

AI Resume Agent Architecture: ┌─────────────────────────────────────────────────────────────┐ │ FRONTEND: Floating Chat (React + Framer Motion) │ │ SparklesChatContext → FloatingAIChat → Streaming Display │ └─────────────────────────────────────────────────────────────┘ ↓ HTTP/SSE ┌─────────────────────────────────────────────────────────────┐ │ API: /api/resume-chat (Next.js API Route) │ │ Session Management → Context Building → Claude 4.5 Opus │ └─────────────────────────────────────────────────────────────┘ ↓ ┌─────────────────────────────────────────────────────────────┐ │ JD ANALYSIS: Intelligent Processing (753 LOC) │ │ Technology Detection → Role Scoring → Match Analysis │ └─────────────────────────────────────────────────────────────┘ ↓ ┌─────────────────────────────────────────────────────────────┐ │ RAG LAYER: Hybrid Search (343 LOC) │ │ Vector (pgvector) + Keyword (ILIKE) + Facts (TypeScript) │ └─────────────────────────────────────────────────────────────┘ ↓ ┌─────────────────────────────────────────────────────────────┐ │ DATA: Supabase │ │ resume_embeddings | resume_facts | memory_sessions │ └─────────────────────────────────────────────────────────────┘