Orbinix

The Problem

Job seekers spend hours rewriting resumes for each application. Generic AI tools fabricate experience, creating credibility risks. Hiring managers reject candidates when they detect inflated claims. The market needed a system that optimizes resumes without crossing the line into hallucination.

The Solution

Orbinix takes a user's complete, honest work history as a master CV, then intelligently rewrites, reorders, and optimizes it to match any job description. The critical constraint: every output bullet must trace back to something the user actually provided. No fabrication. Ever.

• 1. Multi-model routing layer — selects the best LLM for each task (GPT-4o for structure, Claude 3.5 for reasoning, ElevenLabs for voice cover letters).
• 2. Defensively correct billing — every LLM call is tracked at the token level with hard limits and real-time cost visualization.
• 3. Azure self-hosted deployment — not a one-click Vercel deploy. Real VM management, reverse proxy, SSL, and automated backups.
• 4. Headless CMS — the entire marketing site and blog are powered by Contentful, accessed via Delivery API and Preview API in draft mode.

Results

Architecture Highlights

The system is built as a series of composable pipelines. Each pipeline stage is independently testable, versioned, and replaceable. The prompt layer uses a template inheritance system: base constraints are defined once, and job-specific instructions are layered on top without duplication.

Cost controls are not an afterthought. A middleware tier intercepts every LLM request, estimates cost before execution, and enforces per-user and per-session budgets. If a request would exceed the limit, it is queued for manual review rather than silently truncated.