Inspiration

Compliance audits are slow, expensive, and risky. Cloud LLMs expose confidential data. As brothers solving enterprise bottlenecks, we built a tool to collapse a 3-week gap analysis into minutes, keeping every byte 100% local.

What it does

TrustNode is a 100% offline, AI-driven compliance engine. It evaluates corporate policies against standards (ISO 27001, GDPR) and calculates a definitive metric:

$$\text{Trust Score} = \left( \frac{C + 0.5P}{N} \right) \times 100$$

(Where $C$=Compliant, $P$=Partial, and $N$=Total controls). It outputs a publication-grade LaTeX PDF report with heatmaps, gap analysis, and a multilingual AI executive summary.

How we built it

A zero-network-egress system:

Frontend: React 19, Vite, and Tailwind v4 (glassmorphic UI).
Backend: FastAPI.
Local RAG: PyMuPDF (parsing), nomic-embed-text & ChromaDB (vector store).
Inference: Ollama running llama3.1:8b for strict JSON evaluation.
Export: Python-to-LaTeX bridge for dynamic PDF generation.

Challenges we ran into

Running local RAG on Fedora is resource-heavy. Forcing strict JSON from Llama 3.1 required relentless prompt engineering. Also, integrating the LaTeX compiler right before the deadline meant quickly writing a text sanitizer to escape special characters (like \% and \$) to prevent compilation crashes.

What we learned

Orchestrating local vector DBs with local LLMs, and bridging modern web frameworks with academic typesetting (LaTeX) to produce institutional artifacts.