CareerForge

Real-Time AI Chat

CareerForge streams Claude CLI responses in real time over WebSocket. The chat system supports multi-session conversations, mid-stream steering, message queuing, and file attachments. Each session persists to SQLite and can be resumed across browser reconnections.

System Architecture

A full-stack TypeScript application with a React 19 frontend, Express 5 backend, SQLite database, and a custom Claude CLI bridge that spawns and manages AI sessions as child processes with full duplex streaming.

The Claude CLI Bridge

The defining technical challenge. CareerForge spawns claude as a child process, pipes messages through stdin, and parses streaming JSON from stdout in real time. Sessions persist across browser reconnections via "orphaned bridge" management — if a client disconnects while Claude is processing, the bridge keeps running and reattaches when the client returns.

sequenceDiagram
    participant Browser
    participant WebSocket
    participant Bridge as Claude Bridge
    participant CLI as Claude CLI

    Browser->>WebSocket: chat message
    WebSocket->>Bridge: sendMessage(text)
    Bridge->>CLI: spawn claude -p --stream-json
    Bridge->>CLI: write to stdin
    loop Streaming Response
        CLI-->>Bridge: stdout JSON chunks
        Bridge-->>WebSocket: emit "stream" event
        WebSocket-->>Browser: real-time text delta
    end
    CLI-->>Bridge: process close
    Bridge->>Bridge: save to DB + disk
    Bridge-->>WebSocket: emit "complete"
    WebSocket-->>Browser: full response
						

Job Pipeline

A Kanban-style job board tracking 46 opportunities through a 13-stage pipeline — from initial discovery through interviews to offer negotiation. Each job card shows fit score, compensation range, and links to auto-generated artifacts: tailored resume, cover letter, and skill gap analysis.

Skills Assessment

338 skills across 10 color-coded categories, extracted from job opportunities and the master resume. Each skill gets a 1-10 self-rating with button-style selectors. Skills rated 6+ are flagged as resume-ready. Generate Study triggers Claude to create study materials that auto-populate the Study Guide with multiple-choice interview prep questions.

Structured Journal

AI-powered journaling with auto-parsed sections, mood tracking, energy levels, inline hashtag extraction, and principle linking. Journal entries are created through Claude chat sessions in a special "journal mode" that structures freeform thoughts into categorized sections with action items.

Data Architecture

28 SQLite tables across 7 migrations manage the full lifecycle — from job discovery through skill assessment, study generation, journaling, and interview preparation. Three auto-indexers watch the file system for changes and keep the database synchronized.

flowchart LR
    subgraph Indexers["Auto-Indexers"]
        OI["Opportunity
Indexer"]
        SI["Study
Indexer"]
        JI["Journal
Indexer"]
    end

    subgraph FS["File System"]
        OPP["jobs/opportunities/
44 folders"]
        JRN["content/journal/
Markdown files"]
    end

    subgraph DB["SQLite (28 tables)"]
        JOBS["Jobs Pipeline
13 statuses"]
        SKILLS["Skill Assessments
338 skills"]
        STUDY["Study Topics
+ MC Questions"]
        JOURNAL["Journal Entries
Sections + Tags"]
        CHAT["Chat Sessions
+ History"]
    end

    FS --> Indexers
    OI --> JOBS
    OI --> STUDY
    SI --> STUDY
    JI --> JOURNAL

    style Indexers fill:#1a1000,stroke:#f59e0b
    style FS fill:#0a1520,stroke:#22d3ee
    style DB fill:#0a1a15,stroke:#10b981
						

Study Guide Generation

When you click "Generate Study" on a skill, the system creates multiple-choice interview prep questions immediately (adapted to your self-rated proficiency level), then queues a background Claude CLI job to generate comprehensive study material with learning paths, practice exercises, and curated resources.

Feature Inventory

Technology Stack

Key Architecture Decisions