Tribunejustice

The Problem

Finding a lawyer in Cameroon is slow, opaque, and often unreliable. Clients have no visibility on costs upfront, no way to track their case progress, and payments flow through informal channels with no guarantee of security. Lawyers, on their side, struggle to manage their caseload, collect payments, and communicate with clients across fragmented tools.

Tribunejustice needed a platform that would solve this from both ends — bringing trust, structure, and speed to a historically informal market.

Scope — Three Applications, Built End-to-End for a Legal Expert

This wasn't a website. It's a full product platform — the digital product of Mr Badjeu Kuitchouha Ghislain, a Cameroonian legal expert who came up with the idea and wrote the specifications. He consulted me to build it: I led and built the entire technical side — from the data model (MCD) to the architecture, from security hardening to deployment — over 8 months (~1,650 tracked changes), exchanging with him on every product decision as ideas flowed both ways.

What I Built

Domain engine — the Service Request: The entire platform is organized around one central entity carrying the full lifecycle: creation, expert assignment, tracking, payment, resolution. A ServiceRequestStatusMachine guarantees no case can skip a step or revert to an invalid state — every transition is validated and logged. The 55+ models were designed from a single MCD, so the whole product shares one coherent vocabulary.

Video consultations & case tracking: Appointments with availability management, video sessions integrated with the case lifecycle, shared documents, and billing automatically attached to the ServiceRequest — a full consultation workflow, not just a chat widget.

Payments with escrow (MeSomb): Cameroonian mobile money (MTN Mobile Money, Orange Money) via MeSomb, with an escrow flow that holds funds until service completion, professional payouts, transaction tracking, and invoice generation (PDF). Trust is the core product of a legaltech — payments were designed around it.

Real-time communication: Client-expert messaging over WebSockets (Laravel Reverb) with presence, read receipts, and favorites — no polling. Horizon queue workers dispatch web push, email, and SMS notifications asynchronously, so delivery never blocks the request.

Secure document vault: A dedicated vault gives clients a structured, access-controlled home for sensitive legal documents, tied to the case lifecycle.

Search & discovery (Typesense): Expert profiles are indexed with faceted, typo-tolerant search in real time — reducing time-to-match while keeping relevance control in the backend.

CRM & automation: A full CRM module tracks contacts, lifecycle stages, lead scoring, and campaign automation — connected to the core domain via observers and events, so CRM activity reacts to real case events instead of living in a silo.

Frontend architecture — ApiClient: Every API call goes through a centralized client handling exponential backoff with jitter, idempotency keys, in-flight request deduplication, response normalization, and typed errors (NetworkError, ValidationError, ServerError…). The monolithic AuthService (594 lines, 87 consumers) was refactored into 4 signal-based services — state, API, token lifecycle, permissions — making auth predictable and testable.

Bilingual by design (FR/EN): Both official languages of the Cameroonian justice system, across all three applications — with translation tooling, i18n key auditing, and telemetry that reports untranslated strings so gaps are fixed before users see them.

Performance & reliability: LCP went from 5,004 ms to under 1,500 ms. Image assets were converted to WebP — 68 MB → 3.9 MB (94% lighter). Deployments became zero-downtime, stale-cache crashes auto-recover, PM2 keeps processes alive, Redis replaced file cache, and backups are automated. The platform stopped breaking when it was updated.

Security — defense-in-depth: Legal data is highly sensitive, so the platform was hardened through two audit campaigns (external pentest + internal review):

• 41 vulnerabilities closed across the blog, RBAC system, and frontend — before any breach occurred
• HTTP-only JWT cookies, CSP finalized on all 3 domains, CORS deduplication, security.txt
• Sudo mode (password re-confirmation, 10-minute Redis TTL) for sensitive actions: 2FA changes, data exports, admin operations
• Granular RBAC with strict Admin / Super Admin separation; 2FA with recovery codes
• SSRF protection on the blog image proxy via strict allowlist; open-redirect and XSS guards on the frontend
• RGPD compliance in progress: data-processing register, impact analysis, business-continuity and incident-response plans

Impact

Security Audit Results

• 17 blog vulnerabilities fixed: SSRF proxy, GET→POST logout, missing auth middleware, open redirect
• 14 RBAC flaws: Admin/Super Admin separation, critical permission protection (roles, permissions, settings)
• 10 frontend fixes: XSS via isomorphic-dompurify, open redirect via isSafeRedirectUrl(), fail-closed guards
• Sudo mode audit: 2FA enable/confirm routes without protection, change-password without sudo, broken admin verify-email route

What This Experience Taught Me

Leading the end-to-end build of a legaltech platform for a client forces you to own every layer — and it changed how I approach systems. Architecting a legaltech platform forces a specific mindset: every feature has to answer two questions — "can this data leak?" and "can this workflow be subverted?" Security can't be layered on after the fact in a domain handling sensitive legal and financial data.

I also learned to balance velocity against quality: shipping security fixes without breaking user experience, migrating an Angular codebase without halting production, and documenting architectural decisions (state machine, idempotency, sudo mode) so that the platform keeps evolving without accumulating debt.

Next Steps

The platform is now serving as a testing ground for AI agent integration — automated first responses, legal document analysis, and complex workflow orchestration directly connected to production data.