In this article

The work began with an understanding of the problem

I first built PROFAS in 2017 while helping with a fingerprint PhD project. It was the first thing I coded: React, Cerebral for state management and Fabric.js for the canvas. I spent about eight months working on it, generally for at least three hours every day. The code was subsequently lost. What survived was a short progress video.

That video was enough to recover a surprising amount of the interaction: analysis prompts, quality regions, feature positions, comparison relationships and the progression towards an opinion. More importantly, I still knew the problem those controls were intended to address. How does an observation become a record, and how does someone else inspect the reasoning behind a conclusion?

The current build took nine days from creating the first PROFAS Convex project. Those days sit on top of forensic study, the original project and years of industrial software and automation work. Agent-assisted implementation made that accumulated context actionable. The useful engineering account starts with that context and follows the choices that turned it into a system.

Establish the record before widening the tools

The first workspace was canvas-led. A questioned impression and its reference could be inspected with specialist tools close to the material. Reproducing the visible interface was only the beginning. An annotation needed a source, an author and an accepted position in the examination record; a submitted stage needed to retain the work supporting it.

That led into persistence, revision handling and independent review. A verifier’s initial work had to remain distinguishable from the review made after seeing another examiner’s findings. Opening reference material and disclosing a colleague’s conclusion were meaningful events, with different information available before and after them.

Once that structure existed, new tools had a definite place to contribute. Ridge counts, map operations, spoken notes and agent commands could be assessed against the same question: what exactly will this action add to the record, and what basis will remain available afterwards? This reduced the number of independent meanings the interface had to invent.

The examination record retains accepted contributions and their development. This prepared fingerprint example illustrates the product interface.

Build the surrounding systems with clear ownership

PROFAS now has fingerprint, footwear, tyre, firearm and toolmark, and DNA STR modules. Their analytical material differs substantially. Photographs use original image pixels; surface work uses measured grids; DNA observations use channel and scan coordinates alongside separately imported peak analysis. They share a record and review framework while retaining specialist tools and source models.

Around the examination application sit three other active applications: the public website, a private business workspace and a research library. Each owns its runtime, dependencies, configuration and data. The retained fingerprint application also remains separate. Keeping them in one repository makes their design context available during development without requiring them to share a running system.

Institutional delivery applies another boundary. The maintained examination interface can run against a separate backend for each institution. A business service entry describes that relationship; it does not grant access to examination records. This costs more care in provisioning and release selection, but makes ownership explicit.

ENGINEERING NOTE

One project, distinct running systems

  1. Examination platform

    Specialist modules, accepted records, evaluation and controlled review.

  2. Public website

    Product explanation, demonstrations and articles using approved public captures.

  3. Business workspace

    Mail, relationships, tasks and institutional service records.

  4. Research library

    Source-linked papers, reviews, recall material and selected research relationships.

The applications share development context in the repository. They retain independent runtimes, credentials and data ownership.

The revisions belong in the story

Some early decisions changed quickly. Passkey authentication was a poor fit for unattended development and the intended university workflow, so it was replaced with email and password authentication. Public account creation later gave way to invitation-only access. The important change was aligning access with how the platform would actually be introduced and used.

Demo access also became simpler. Prospective users needed approved example pairs they could open and explore. Requiring additional setup before an examination created resistance at precisely the wrong point. The demonstration workflow was revised around starting an example with minimal preparation.

The research library followed a similar course. A local ParadeDB implementation supplied search, source reading and portable exports. When shared online access became the requirement, SQLite FTS5 in D1 was sufficient for the actual retrieval task. Private originals moved into R2 for the hosted library, while the local installation remained available for extraction and recovery. Choosing a smaller requirement changed the infrastructure decision.

What nine days can tell us

The repository contains working interfaces and the rules behind them, supported by dated delivery records. Those records distinguish source inspection, focused behavioural checks, browser observations, isolated recovery rehearsals and publication. A successful build answers a different question from a restored examination retaining its original source basis. Keeping those distinctions clear made the result easier to assess.

The development method evolved alongside the product: one continuing coordinator conversation, usually two bounded specialist assignments, explicit ownership and integration into a known source base. The most useful acceleration came when a task had a stable contract and a concrete example. Shared decisions still needed to be made, and generated changes still needed to be understood.

This is one build, rather than a measured comparison between development methods. Its interfaces and recording rules provide infrastructure for research; studies and institutional use must establish their own findings. The engineering contribution worth exploring is already concrete: a common record, native evidence models, separate operational boundaries and a development process that kept those choices connected as the scope grew.