---
title: Veridian LQIS Demo
emoji: 📉
colorFrom: indigo
colorTo: red
sdk: docker
app_port: 7860
pinned: false
license: mit
short_description: Explainable deal-loss prediction with live coaching
---

# Veridian.ai — Lost Quote Intelligence System (LQIS) · Live Demo

Paste (or build) a buyer–seller negotiation turn by turn and watch the **loss probability** update
live, with the **turn where the deal tipped** marked on the trajectory and **evidence-cited coaching**.

- Model: [`sumitnewold/veridian-lqis`](https://huggingface.co/sumitnewold/veridian-lqis)
- Hierarchical RoBERTa → Temporal Transformer → **FiLM** fusion → calibrated loss probability + SHAP coaching.
- **Methodology demonstration** on Craigslist (C2C) bargaining — generic/non-negotiation chat is
  out-of-distribution and reads near 0%. External market line is demo-only (not used in the trained score).

Free CPU Space — first request loads RoBERTa (a few seconds per turn). Built from a Docker image
that serves the FastAPI backend (`/score`) and the static site at `/`.