Skip to content
From crash scene to signed verdict
Forensic LegalTech · Accident Reconstruction
Web & App Development AI & Automation Integrations

IRS VIAL

From crash scene to signed verdict

IRS VIAL has reconstructed traffic accidents in Colombia for 19 years. Its reports were written in Word, coordinated over WhatsApp and carried no digital legal weight. We built IRS Móvil: a forensic platform where the client opens the case and uploads evidence, AI drafts the report, the forensic expert edits it and the director signs it digitally with full legal validity in Colombia — with a SHA-256 chain of custody anyone can verify.

Duration

Ongoing · since June 2026

Started

June 9, 2026

From crash scene to signed verdict
8
Systems in production

Client portal, expert workflow, sketch editor, signing, AI, custody, payments, backoffice

10
AI touchpoints

Across the forensic flow, always with a human in the loop

PAdES
Legally valid signature

Certicámara certificate, full legal validity under Law 527/1999

74
Database migrations

~81k lines of TypeScript, 233 commits in about four months

The Challenge

IRS VIAL is a Colombian family firm with two decades of accident reconstruction behind it and a large archive of reports in paper and Word.

Its earlier mobile app had been withdrawn for low adoption, and the forensic work itself — receiving evidence, analyzing it, classifying the cause, drafting the Technical-Legal Concept (CTJ) and signing it — happened outside any system. More than 60% of Colombian vehicles have no voluntary insurance, and their owners need a solid technical opinion before settling or going to court.

  • — A forensic workflow with no system: no case status, no work queue, no standard deliverable
  • — No digital legal validity: no electronic signature and no chain of custody to defend a report before an insurer or a judge
  • — Two decades of reports in paper and Word, and a mobile app withdrawn for low adoption
  • — Strict regulation: NTC 3701 causality, Resolution 11268 hypotheses and Law 1581/2012 data protection, all to be modeled, not improvised
Our Solution

IRS Móvil, an end-to-end forensic platform with eight systems in production.

A case moves from client intake to expert analysis to the director’s digital signature and delivery, and every piece of evidence is sealed along the way.

  • + Client portal · a digital garage for vehicles, drivers and documents (SOAT, inspections, policies), a 6-step case wizard with SOAT OCR, voice-recorded statements with AI transcription and consent, and a prepaid credit wallet
  • + Forensic expert workflow · case queue, CTJ editor with editable accident, vehicle and driver data, third-party statements, requests for additional evidence and autosave
  • + Hybrid sketch editor · the expert composes the scene from a curated symbol catalog over satellite imagery; AI seeds the vehicle positions from what the case states, never more
  • + Director review and PAdES signature · approve or return to the expert, then sign with a Certicámara certificate and an institutional seal the client manages without code
  • + Forensic AI assistant · 10 Gemini 2.5 touchpoints: photo quality, transcription, narrative analysis, draft report, accident sequence, NTC 3701 conclusion, photo descriptions, and a sketch that a second model verifies
  • + Chain of custody · every file sealed with SHA-256 and made immutable by database triggers; a public /verificar route lets insurers, lawyers or judges validate a report without an account
  • + Payments and backoffice · Bold payments credit the wallet, a service marketplace (IVAT, reconstruction), exportable reports, configurable notifications and an editable CMS for the landing and login
Selected Frames

Inside the build.

Brand, interface and system — shown as they shipped.

Evidence upload step with AI photo-quality scoring
CLIENT WIZARD · EVIDENCE
Visual evidence guide, AI photo-quality score and SHA-256 sealing on upload
IRS Móvil on mobile
MOBILE
The client side works on the phone, where accidents happen
Forensic expert CTJ editor with AI suggestions
EXPERT EDITOR · CTJ DRAFTING
AI proposes the sequence and hypothesis, each claim linked to its evidence; the expert accepts, edits or regenerates
Accident scene map with impact point and trajectories
SCENE · IMPACT POINT
Exact location, impact point and each vehicle’s trajectory
Director review and signature
DIRECTOR · PADES SIGNATURE
Final review with an AI checklist, then a legally valid digital signature
Public SHA-256 verification page
PUBLIC VERIFICATION
Anyone can check that a piece of evidence has not been altered
How the AI works in the forensic pipeline
AI PIPELINE
AI speeds up the expert; it does not replace them
Under the hood

The backend is the product.

IRS Móvil is not a website. It is a forensic system with a regulation-shaped data model, business rules that carry legal weight, custody cryptography and domain AI. By our estimate about 60% of the effort is backend, behind 84 screens split across five roles.

31
Postgres tables, 30 with row-level security
52
Row-level security policies
33
Database triggers
31
SQL functions
61
Server-action modules
84
Screens across 5 roles
01

Ten AI touchpoints, zero autonomous decisions

Prompts are built from the real case — the exact vehicle roster, damage and hypotheses — with hard “no invention” constraints and per-case cost caps. A second model checks every generated sketch. The expert edits, the director signs; the AI never closes a case.

02

A chain of custody anyone can check

Each file and audio is hashed with SHA-256 and locked by database triggers. A public token is generated when the case is submitted, and the /verificar route lets a third party validate the report and download the custody package without access to the platform.

03

Legally valid PAdES signatures

Reports are signed with a Certicámara .p12 certificate in PAdES format, with the full CA chain and an institutional seal. If the certificate fails, the system falls back to a hash-based electronic signature so delivery never breaks.

04

A legal PDF that stays faithful

Two synchronized renderers — the signed PDF and an HTML preview — with justified text, NTC 3701 tables and photos resized so the signed report stays within storage limits.

05

Large evidence, direct to storage

Photos, audio and site plans upload straight from the browser through signed URLs, past the 4.5 MB serverless limit, then get validated by magic bytes and sealed with their custody hash on the server.

06

Hybrid AI Coding

A senior WebMinds engineer directs AI agents (Claude Code) that implement, and an independent QA agent reviews every change before production. That is how 233 commits and 74 migrations shipped in about four months, across 37 deploy days.

The Stack

Technologies & Tools

The tech stack powering this solution

App
Next.js
UI
React
Postgres · RLS · Vault
Supabase
Forensic AI
Gemini 2.5
Digital signature
PAdES
Payments
Bold
Geolocation
Google Maps
Email
Resend
QA
Playwright
Claude Code
Hybrid AI Coding
The Impact

Beyond the numbers, we delivered a system that transformed how this client operates.

01

A 19-year forensic practice, now software

From Word documents and WhatsApp to a platform that takes an accident from the scene to a digitally signed report with legal validity and a chain of custody third parties can verify.

02

Domain AI, humans in control

Ten AI touchpoints speed up the expert — photo review, transcription, drafting, sequence, NTC 3701 classification, sketches — with anti-hallucination guards and the expert and director always in charge.

03

The client runs it without code

Digital signature, pricing, services, notifications and even the landing and login content are managed from the backoffice. We handed over autonomy, not dependency.

Timeline
  1. Jun 9, 2026
    Project kickoff: Next.js, Tailwind and the design system
  2. Jun 10, 2026
    Database with row-level security, NTC 3701 catalog, roles and dashboards
  3. Jun 16, 2026
    Full client flow: 6-step wizard, wallet, cases and public verification
  4. Jul 2026
    Expert workflow, CTJ editor and director PAdES signature, end to end
  5. Aug 2026
    Hybrid sketch editor, OCR, Bold payments and reports
  6. Sep 2026
    Certicámara signature in production, configurable notifications, direct-to-storage uploads
  7. Next
    Certificate activation, pilot data cleanup and end-to-end verification tests
Start Your Project

Ready to achieve similar results?

Let's discuss how we can transform your business with a tailored digital solution.

Hi! I'm Atlas