---
session_id: "1381580"
title: "The first Grounded Decision Model"
phase: live
last_checked: 2026-11-12T18:52:58-08:00
render: 20261113-025258-ddff00
---

# Links: The first Grounded Decision Model

## Speaker

**Devin Gupta**, Core engineer, Decisions API, Parallel Web Systems

- Site: [Devin Gupta](https://devin-gupta.github.io/)
- X: [@devingupta_](https://x.com/devingupta_)

## Company

**[Parallel Web Systems: Decisions API](https://parallel.ai/products/decisions)**:
"A decision engine with world knowledge built in. Typed answers grounded in live
web data, with citations, in under a second. $1 per 1,000 requests." (from the
site)

## Resources

- **Blog** · [Introducing the Parallel Decisions API](https://parallel.ai/blog/decisions-api)
  (Parallel). The launch post for the product the talk demos ("since we launched
  two weeks ago"): several typed questions about one entity in a single grounded call.
- **Blog** · [Introducing PWS-Decisions: A benchmark for web-grounded decisions](https://parallel.ai/blog/pws-decisions-benchmark)
  (Parallel). The open 300-question benchmark and the grounded-versus-ungrounded
  accuracy gap.
- **Repo** · [parallel-web/pws-decisions-benchmark](https://github.com/parallel-web/pws-decisions-benchmark).
  Run the benchmark against your own model.
- **Dataset** · [parallel-web/pws-decisions-benchmark](https://huggingface.co/datasets/parallel-web/pws-decisions-benchmark)
  (Hugging Face). The 300 questions.

## Gaps

- His LinkedIn (from the roster) is login-walled and returned no title, so it is
  not linked.
- Neither Parallel post names an author.
- The abstract's cost figures (104 vs 4,777 tokens; model share 86% to 17%), the
  4,270-business demo and the entity-resolution failure have no write-up found.
- The ASR is heavily garbled; the recording opens mid-sentence.
- No official recording or slides found yet. The dossier rechecks every 6 hours
  during the event.