Skip to content

prls

For b2b AI companies to measure their visibility in LLM.

Code Video Cerebral Valley

Core trick

Masks the provider name in a transcript-derived brief, then checks whether GPT-5 and GPT-4o still infer the masked company.

Tags

productivityagents

Description

For b2b AI companies to measure their visibility in LLM. Generate brand-visibility briefs by masking provider names and comparing how multiple LLMs (GPT-5 primary, GPT-4o comparison) respond. The pipeline ingests a transcript, extracts selling points, asks domain-expert questions, and highlights where models still infer the masked provider.
JessieJessJe/story-ai-visibility★ 10 forksPythonCreated Sep 27, 2025 · last push Sep 28, 2025

README excerpt

# Story AI Visibility Agents

Generate brand-visibility briefs by masking provider names and comparing how multiple LLMs (GPT-5 primary, GPT-4o comparison) respond. The pipeline ingests a transcript, extracts selling points, asks domain-expert questions, and highlights where models still infer the masked provider.

## Judges, here’s the TL;DR

- Live deployment: `https://story-ai-visibility-fe.vercel.app`
- **Codex in Development**: `AGENTS.md` defines agent roles, while `docs/PRD.md` and `docs/PRD_v2.md` specify the delivery roadmap; the Codex CLI tool enforces those specs by running pytest, linting prompts, and shipping atomic patches so every change is reviewable.
- **GPT-5 in Project**: `src/cli.py` drives GPT-5 through `assets/visibility/system.prompt.md`, `assets/visibility/extract_pillars.prompt.md`, and `assets/visibility/generate_questions.prompt.md` to mask providers, generate probes, and score inference; GPT-4o cross-checks outputs, and stub fixtures keep evaluations reproducible.
- **Live Demo**: TBD — planned demo highlights side-by-side GPT-5 vs GPT-4o inference detection, latency, and masking compliance once validation logging is finalized.
- **Technicality**: 1) Algorithm — deterministic ingestion → provider masking → pillar extraction → question generation → inference scoring with confidence flags; 2) Tech stack — Python 3.11+, FastAPI, modular `src/common` utilities, pytest coverage targets ≥85%, Codex-driven scripts under `scripts/`.
- **Frontend Notes**: R

Team