VISION × TIME × REASONING

THE PHYSICAL WORLD IS NOT A STILL IMAGE.

Reality happens over time.

Duranta builds vision-first multimodal intelligence for 4D reasoning — understanding objects, motion and sequence as one continuous event.

Explore the claim workspace ↘ See how it works
OBSERVATION 14:02:11 EVENT VECTOR EVIDENCE TRACE / 04
01 / THE PREMISE

Frames show evidence.
Sequences add context.

Most AI systems flatten the world into isolated images, text fields and labels. But a claim is an event. Understanding it means preserving the relationship between what was seen, where it happened, when it changed and what followed.

01

See the whole event

Video, images, documents, sensor traces and human statements are aligned against a shared timeline.

02

Reason in context

Signals become spatial and temporal relationships, not disconnected labels passed between separate tools.

03

Show the work

Each sample interpretation stays connected to the underlying observation, so a human can inspect the chain of reasoning.

02 / CLAIM WORKSPACE

One event. Every signal in view.

Move through an illustrative collision sequence. The scene, evidence state and event interpretation advance together instead of forcing an adjuster to reconcile disconnected tabs.

FRAME GROUP / T−03.2s

Independent motion established

Both objects are tracked against the same scene geometry and event clock.

OBSERVATIONS6 aligned
STATEpre-event
REVIEWhuman required
03 / BETA PRODUCTS

Four beta products.

Open each beta product to inspect its synthetic scenario, local-file path and review boundaries. The product demos also use publicly available records where sourced.

P / 01

Accident Replay

Vehicle collision reconstruction from event data and telemetry.

Open beta product ↗
P / 03

Liability Replay

Venue incident reconstruction across camera and operations logs.

Open beta product ↗
04 / DEMONSTRATIONS

Six browser demos.

Explore synthetic scenarios and sourced, publicly available records, or load a local file. The records do not supply the invented replay sequences; these are browser demonstrations, not live claim assessments. Files stay in this browser.

D / 01

Accident Replay

Vehicle collision reconstruction from event data and telemetry.

Open demo ↘
D / 02

Property Replay

Property damage sequence from sensor and inspection evidence.

Open demo ↘
D / 03

Liability Replay

Venue incident reconstruction across camera and operations logs.

Open demo ↘
D / 04

Claim Ready

Site incident records prepared for human claim review.

Open demo ↘
D / 05

Storm Replay

Synthetic catastrophe track and portfolio exposure.

Open demo ↘
D / 06

Fleet Replay

Synthetic multi-vehicle incident cascade.

Open demo ↘
D / 01
SYNTHETIC SCENARIO — GENERATED SAMPLE

Accident Replay

Explore an example collision sequence from generated event data. The public-record scenario cites crash-level fields, not measured telemetry or video.

1 Files stay in this browser.

Select Run synthetic scenario

PHASE / ——

The event record will appear here.

SPEED / MOTION—
VEHICLE SIGNAL—
CONTEXT—
0 / 0

Duranta shows what the evidence says and does not decide fault, liability, or coverage — people do.

D / 02
SYNTHETIC SCENARIO — GENERATED SAMPLE

Property Replay

Explore a generated damage sequence. The public-record scenario cites claim-level fields, not measured sensor or inspection events.

2 Files stay in this browser.

Select Run synthetic scenario

PHASE / ——

The event record will appear here.

WATER / SMOKE—
TEMPERATURE—
DOCUMENT TRACE—
0 / 0

Duranta shows what the evidence says and does not decide fault, liability, or coverage — people do.

D / 03
SYNTHETIC SCENARIO — GENERATED SAMPLE

Liability Replay

Explore a generated venue sequence. The public-record scenario cites crash-level fields, not footage or venue logs.

3 Files stay in this browser.

Select Run synthetic scenario

PHASE / ——

The event record will appear here.

CAMERA EVENT—
OPERATIONS LOG—
HUMAN RECORD—
0 / 0

Duranta shows what the evidence says and does not decide fault, liability, or coverage — people do.

D / 04
SYNTHETIC SCENARIO — GENERATED SAMPLE

Claim Ready

Review synthetic sample incidents across sites. Each generated incident appears as a sample event record with event metadata only.

4 Files stay in this browser.

Select Run synthetic scenario

PHASE / ——

The sample incident record will appear here.

SITE—
RECORD STATE—
EVIDENCE POLICY—
0 / 0

Duranta shows what the evidence says and does not decide fault, liability, or coverage — people do.NO VIDEO LEAVES THE SITE.

D / 05
SYNTHETIC SCENARIO — GENERATED SAMPLE

Storm Replay

Catastrophe event track and portfolio exposure.

Generated scenario. Does not interpret uploaded video into claim facts or make fault, liability, or coverage decisions.

5 Files stay in this browser.

Select Run synthetic scenario

PHASE / ——

The event track will appear here.

LOCATION—
EVENT / MAGNITUDE—
AFFECTED EXPOSURE—
0 / 0

Synthetic track and exposure only. Illustrative sample; no claim decisioning.

D / 06
SYNTHETIC SCENARIO — GENERATED SAMPLE

Fleet Replay

Commercial fleet incident cascade across vehicles.

Generated scenario. Does not interpret uploaded video into claim facts or make fault, liability, or coverage decisions.

Files stay in this browser.

Select Run synthetic scenario

PHASE / ——

The fleet cascade will appear here.

VEHICLES 01–02—
VEHICLES 03–04—
VEHICLE 05 / CASCADE—
0 / 0

Synthetic fleet telemetry only. Illustrative sample; no fault, liability, or coverage decisions.

05 / ARCHITECTURE

A reasoning layer built around events.

The product direction is to connect raw multimodal evidence into a navigable event model, giving operators a precise place to investigate, challenge and act.

Ingest

Bring together visual, textual and sensor evidence without losing source or sequence.

Align

Map observations into a shared spatial-temporal representation of the event.

Reason

Evaluate relationships, contradictions and candidate explanations across the full sequence.

Review

Present candidate interpretations with the evidence trace a reviewer needs to understand and challenge them.

06 / CASE STUDIES

From fragmented evidence to one inspectable record.

These synthetic scenarios illustrate how each reconstruction keeps observations, timing and uncertainty connected for human review.

SYNTHETIC / ILLUSTRATIVE · ACCIDENT REPLAY

Wet-road intersection collision

EVENT

Two vehicles converged through a signalized intersection during light rain. Braking began 1.1 seconds before observed contact.

EVIDENCE USED

Dashcam event frames, vehicle speed and brake telemetry, GPS movement, weather and road-surface observations.

RESULTING RECORD

A second-by-second sequence connecting approach, braking, contact and rest positions, with each statement linked to its source.

SYNTHETIC / ILLUSTRATIVE · PROPERTY REPLAY

Multi-floor water damage

EVENT

A supply-line pressure drop preceded moisture alarms in two rooms and visible ceiling damage on the floor below.

EVIDENCE USED

Water sensors, temperature readings, site photos, maintenance video and a contractor inspection document.

RESULTING RECORD

A minute-by-minute reconstruction of first signal, spread, isolation and inspection, with observed and inferred states separated.

SYNTHETIC / ILLUSTRATIVE · LIABILITY REPLAY

Venue entrance fall

EVENT

A guest fell near the entrance shortly after a floor service task and before an incident report was opened.

EVIDENCE USED

On-site camera events, door access logs, till timestamps, staffing records and the signed incident report.

RESULTING RECORD

A synchronized venue timeline showing the floor condition, staff movements, guest entry and documented response for review.

FOR CLAIM REVIEW

Claims are where fragmented evidence becomes an expensive decision.

Reconstruct what happened, expose uncertainty and keep a human in control of the outcome.

01
Reconstruct the event before judging the claim.
02
Separate observed fact from model inference.
03
Make uncertainty inspectable, not invisible.

ONE EVENT, MANY SOURCES.

Give operators a model of what happened — not another pile of inputs.

SOURCE NOTES

What the public records establish.

  1. NHTSA FARS 2023, ST_CASE 10019: crash-level place, time, collision coding and one fatality; no video or second-by-second vehicle telemetry. Source CSV.
  2. FEMA NFIP Redacted Claims v3, id 6320555 (v2 id 3296480f-ddf8-4264-abd7-aa51b2c57ae4): loss date, location band, flood zone and payment/damage fields; no sensor or inspection sequence. The earlier v2 endpoint is scheduled to be retired October 15, 2026; this record is cited from v3. Source API.
  3. NHTSA CRSS 2023, CASENUM 202304637555: coded crash setting, pedestrian event and suspected-serious-injury category; no camera video or exact event chronology. Source CSV.
  4. Texas DOI closed consumer complaints, complaint 467922: received/closed dates and Confirmed finding; no claim file, policyholder narrative or evidence sequence. Source record.
  5. NOAA Storm Events, EVENT_ID 10334009: F0 tornado near Alvo, Cass County, NE, May 8, 1995 at 21:46 CST, reported start 40.87,-96.43 and reported length 0.1 mi. No end coordinates or distinct end minute. Source CSV.

The browser replay frames beyond these public-record fields are generated examples. They are not observations from the cited agencies and do not decide fault, liability, coverage or cause.