DASGPT — Give DAS Data a Voice

A conversational AI agent that turns kilometer-scale fiber-optic sensing (DAS) data into bridge-health answers anyone can act on — a verdict, the reasoning, and a report, in seconds.

Team GeoWhisper  ·  DFW 2026 AI & Startup Competition  ·  Final Submission  ·  July 2026

Open the live demo Watch the 4:43 demo film
QR to live demo

Live demo — full dashboard + assistant on real data. Best on desktop. One-click Chinese/English switch, top-right.

Demo Video

One complete user journey, 4 minutes 43 seconds

Dashboard tour → plain-language question → verified anomaly analysis → standard figures → one-click Word report → publish back to the live screen. Every AI answer on screen is a live backend response — nothing is staged.

If the player shows an error, keep this HTML in the same folder as final.mp4 and final_zh.mp4 (subtitle sidecars final.srt / final_zh.srt included).

The Problem

Bridges talk. Nobody can hear them.

“We have the data — we just can’t read it.”

The Solution

Ask the bridge, in plain language

  • One question in, a verdict out. “Any warnings on the bridge right now?” returns a health verdict, the reasoning behind it, and the exact location to inspect — in seconds, not days.
  • Lives inside the operator’s real dashboard. Embedded in a production bridge-monitoring screen, beside the panels operators already use — with two-way integration: it reads the live alarm state and publishes verified analyses back, with provenance.
  • Standard engineering output. PSD / modal frequencies, strain RMS, anomaly localization, five-level alerts with transparent thresholds, standard figures, and a one-click archivable Word report.
  • Made for non-specialists. The entire experience — dashboard, assistant, reports — switches between Chinese and English in one click.
Dashboard

The live monitoring dashboard with the DASGPT assistant embedded (center), on real bridge data.

Why You Can Trust It

The AI orchestrates; validated algorithms compute

Architecture

The LLM only routes questions to pre-validated engineering code — it never sees raw waveforms and never invents numbers. Every answer is traceable: which algorithm, what parameters, which data.

Not a Mockup

Running live today — on real client data

  • A real client entrusted us with their bridge. Raw DAS recordings (.h5) from an in-service railway bridge: 6 recordings, 1,034 channels along the span, 50 Hz. It is the product’s sole built-in dataset.
  • Every answer is computed live. PSD + modal frequencies, strain RMS, anomaly localization — validated algorithms run on the raw data at question time. No canned answers anywhere in the demo.
  • Public demo for the judges. The button at the top of this page opens the full experience.
Live PSD answer

A live answer: the reference-channel PSD curve with the bridge’s modal peaks, computed on the client’s recordings and drawn directly in the chat.

Customer Validation Evidence

Four trust signals, all within the last two weeks

Everything below is real and verbatim, translated from the original Chinese WeChat threads. Names are redacted; unredacted threads and a 756 KB written review document are available to the judges on request.

1 · A client entrusted us with their bridge

Raw recordings from an in-service railway bridge, handed over for us to build on — a higher-cost trust signal than any survey answer. Now the sole demo dataset.

2 · A senior executive reviewed, then escalated

A senior transport-industry executive tried the public demo the same week we sent it — then forwarded it to his chief engineer and two deputy chief engineers in charge of bridges.

3 · Deployment-grade expert requirements

A six-point written review anchored on our actual demo bridge, plus a 756 KB optimization document from a second reviewer — feedback about a system they expect to run.

4 · Hands-on trials in an industry cooperation

Engineers of a major engineering company (Phase-2 cooperation with SUSTech) tested the live demo — and asked us to record the very demo film in this submission.

“I took a look at the intelligent bridge-health monitoring system this afternoon — the technology is very advanced, the functionality is powerful, and the value for money is good. 👍👍 I asked our two bridge deputy chief engineers to review it as well; I’m forwarding their comments verbatim.” — Senior executive, transport-engineering organization (translated)
“Tried it briefly — works great. I’ll test it more at work.” — Engineer, partner engineering company, after a hands-on trial (translated)

The six-point expert review (summarized) — validated demand, now our roadmap

  1. Auto-correlate strain-peak events with heavy-train / overweight-vehicle passage windows (traceability).
  2. Ingest vehicle-passage, overload and maintenance records into the AI knowledge base so alert thresholds keep correcting themselves — “a self-evolving system”.
  3. Monthly heavy-traffic report: overload passage windows vs. strain peaks — quantitative basis for enforcement and fatigue assessment.
  4. Cumulative fatigue-damage computation for girders and piers, year over year, supporting inspection and major-repair planning.
  5. Season-aware expansion-joint alarm thresholds for the demo bridge’s climate (avoid winter false-alarm storms).
  6. Annual temperature-gradient comparison separating seasonal thermal deformation from permanent structural deformation.
Exhibit A

Exhibit A — the executive’s verdict (“technology very advanced … value for money good”); he asks two deputy chief engineers to review.

Exhibit B

Exhibit B — from the deputy chief engineer’s six-point written review (excerpt; full thread available).

Exhibit C

Exhibit C — partner engineer after a hands-on trial: “works great; I’ll test it more at work.”

Exhibit D

Exhibit D — the partner group asks for a demo video — the film above answers that request.

What we do not claim: no paid contracts yet; structured operator interviews are the first post-competition step. The evidence here is early but real — a senior reviewer, written expert requirements, hands-on trials, all on the live product.

Business Model

Software economics on infrastructure budgets

Pricing and packaging to be calibrated in the first pilot conversations.

Roadmap

From demo to deployment

Next 90 days

1. Formal end-user interviews — bridge operators and government supervisors, run on the live demo.  2. From looped recordings to live streams — real-time ingestion and cloud storage for 24/7 monitoring.  3. First 1–2 pilot bridges, with alarm thresholds and report templates calibrated on site.

Real-time

From replayed recordings to live data streams and cloud storage — 24/7 online monitoring on the same conversational entry point.

Multi-modal

Same fiber, more scenarios: road-subgrade health, traffic flow and vehicle-speed recognition, dynamic vehicle weighing (weigh-in-motion), dam and embankment safety, slope stability, tunnels and underground pipe networks — toward one integrated road–bridge–tunnel–slope entry point. Grounded in real field work: on the Chicao G508 highway, DAS has already done subgrade imaging, traffic-event detection, and vehicle-weight estimation (most vehicles within about one tonne of error).

Engineered for the field

Standardized alarm thresholds and report templates, calibrated per structure against real operating conditions.

Built to scale

Multi-bridge, multi-project deployment with unified data management — from one bridge to a fleet, then tunnels, pipelines, and rail corridors.

Customer-driven backlog: the six-point expert review in the Validation section above is already scheduled into this roadmap — traceability against traffic records, a self-evolving knowledge base, cumulative fatigue-damage tracking, season-aware thresholds.

These are planned directions, stated as such — everything demonstrated elsewhere on this page ships today.

Traction

From thesis to a judgeable product in five weeks

CP1 · Jun 6–7

Thesis + first prototype: conversational analysis on sample data.

CP2 · Jun 29–30

Our team’s production bridge-monitoring platform came together with DASGPT embedded inside — one integrated product, end to end.

Final · Jul 10–12

Real client data end-to-end and made the sole dataset; public deployment; two-way dashboard write-back; one-click Word reports; 4:43 demo film in three cuts; full product manual.

One-click report

A formal monitoring report, generated and downloaded from one sentence.

Team & Ask

AI engineering × DAS domain science, in one team

Mengjia

AI architect — agent harness, product, integration & deployment.

Jiachen Zhong (Jason)

AI architect — agent tooling and evaluation.

Miao Yu

DAS domain expert — bridge algorithms (PSD, modal analysis, anomaly detection).

Pengchao He

DAS domain expert — algorithm development and validation.

Backed by real-world skin in the game: a client provided raw recordings from their in-service bridge, and industry engineers are already testing the live product hands-on.

Our ask

  • Introductions to bridge owners, highway/rail operators, and supervising agencies.
  • Pilot partners — we will run a free pilot on one real structure to calibrate thresholds and reports.
  • Feedback & mentorship from judges with infrastructure or AI go-to-market experience.