Skip to content

Navox Labs

Research first. Build second. Ship when it's real.

In July 2026 we shipped our first fine-tuned model — inkling-hire, built and audited in public in ten days.


What we're building

Product What it does
🧠 inkling-hire Self-hosted AI résumé screening with an always-on bias audit. Powered by a fine-tuned Inkling model (nahrin/inkling-hire) — calibrated scores, blind review by default, human-in-the-loop by construction. First model in the inkling-* series. Your candidate data never leaves your infrastructure.
🔬 Winnow An open, controlled harness for testing bias in AI résumé screeners. Built to find bias; tested four kinds across two production models with real controls — and found they don't robustly replicate on current models. Honest null, open code. The story →
🤖 agents 15 specialist AI agents for Claude Code — full sprint cycle from strategy to shipping. No platform. No login. Your code never leaves your machine.
🕸️ network Map your professional network. Find your side door. Built on 50 years of weak-ties research — no login, no server, your data never leaves your browser.
📊 github-traffic-agent AI-powered GitHub Action that collects repo traffic data daily and generates Claude-powered intelligence briefs. Deterministic data spine with LLM-generated insights — runs on a schedule, no server needed.
🪪 ProfileCard (private beta) Turns your resume into an interactive AI-powered profile card in 60 seconds. Recruiters chat with your AI agent instead of a screening call. Includes voice cloning, profile analytics, AI interview coaching, and a LinkedIn network visualizer.

Research

The work is built on two papers:

📄 The AI Recruitment Crisis — why algorithmic hiring is broken
📄 The Invisible Network — weak-ties theory and the hidden referral economy


The thesis

The systems people depend on — hiring, networking, building software — are broken by design.

Every tool we build is an attempt to give individuals back what those systems took from them.

No VC. No platform. No data stored. Building in public.

Pinned Loading

  1. inkling-hire inkling-hire Public

    Self-hosted AI résumé screening with a built-in bias audit. Fine-tuned Inkling model, calibrated scores, blind review, human-in-the-loop. Your data never leaves.

    Python

  2. winnow winnow Public

    Open, reproducible harness for testing bias in AI résumé screeners (LLM hiring) — counterfactual controls, multi-model, with honest null results.

    Python

  3. agents agents Public

    AI agents for Claude Code — 15 specialists that run your full software development sprint cycle. Strategy, architecture, code review, QA, security audit, and shipping. Open-source Claude Code plugin.

    Python 14 2

  4. network network Public

    Map your professional network. Find your side door. Built on 50 years of weak-ties research — no login, no server, your data never leaves your browser.

    TypeScript 4

  5. github-traffic-agent github-traffic-agent Public

    AI-powered agent that collects GitHub traffic data, reads it with Claude, and sends you a phone-sized brief you can act on. Solves the 14-day retention limit with daily collection, AI analysis, fee…

    Python

  6. pipewar pipewar Public

    Factorio meets cybersecurity. Build production pipelines, defend against hackers. Built entirely by AI agents.

    Python

Repositories

Showing 9 of 9 repositories

Top languages

Loading…

Most used topics

Loading…