Marcio De Melin

Visibility. Orchestration. Human judgment.

AI systems builder for high-trust operational work.

I build proof-first systems shaped by frontline public transport experience: tools that make operational signals easier to capture, workflows easier to coordinate, and high-stakes judgment easier to review without automating away human accountability.

Rapid Report repo LinkedIn Download CV

Why this work is different

Built from operational reality, then generalised carefully.

I am an AI systems builder focused on institutional safety, human-in-the-loop control, and high-trust workflow design.

My work is grounded in 10 years of frontline operational experience at Transport for London, where fragmented information and poor handoffs can quickly become safety, service, and accountability problems.

I do not build chatbots as an endpoint. I build workflow and governance layers: systems that help people capture better signals, preserve boundaries, and make decisions more reviewable.

Rapid Report, Meridian, and Aegis are connected by one question: how can AI make high-pressure institutions more legible without automating away human responsibility?

Project stack

Three connected layers of proof.

Visibility layer

Rapid Report

Rapid Report is a sanitised public demo of an offline-first incident logging PWA for public transport-style operations. The original non-AI version was built independently from frontline observation and used to test whether fast incident capture could improve reporting clarity. The public version is independent, sanitised, and not an official transport authority product or production system.

The demo shows how a single-thumb-tap workflow can turn fast frontline logging into structured shift intelligence, including pattern summaries and a separate local-only AI-assisted summary view. It avoids naming internal systems and keeps the public artifact separate from any operational environment.

Sanitised public demo Independent project Not an official transport authority product
View Rapid Report public repo
Rapid Report home demo
Structured shift summary output
Separate local-only AI-assisted summary layer
AI workflow handoff layer

Meridian — AI Workflow Handoff Layer

Meridian is a configurable AI workflow prototype for high-trust service businesses. It shows how sanitised enquiries can become bounded replies, review notes, missing-detail prompts, follow-up queues, simple reports, and human-owned handoffs.

Meridian demonstrates a human-in-the-loop service workflow pattern: customer signal → bounded reply → missing details → review note → human owner → next step. It is designed to make AI-assisted workflows more legible rather than silently automating decisions.

Local/static demo and proof package complete. Demo data only; not production-deployed, not clinically validated, not stakeholder-approved, and not connected to WhatsApp, CRM, Booksy, or booking systems. No real customer data is used.

Local/static proof package Demo data only Private source remains private
  • Bounded reply drafts
  • Missing-detail prompts
  • Review notes
  • Follow-up queues
  • Simple insight summaries
  • Human-owned handoffs
  • Sanitised reviewer/proof inspector
  • Clear mock/live/demo/production boundaries

Full source and private stakeholder demos remain private while validation is pending.

Meridian desktop proof hero
Customer signal to human-owned handoff
Sanitised reviewer demo
Reviewer handoff package
Decision-support layer

Aegis

Aegis is a human-in-the-loop decision-support prototype for high-stakes operational tradeoffs. It focuses on category-shift recognition, role allocation, escalation clarity, and next-step sequencing.

The prototype explores how fragmented signals and competing priorities can be organised into clearer reasoning without removing human responsibility. Aegis is a design/prototype artifact, not an emergency-response, medical, or operational control system.

Human-in-the-loop prototype No operational control No raw real incident details
Main page
Immediate actions view
Reasoning memo view

Research direction

AI support for accountable human judgment.

My current direction uses public transport-style operations as the starting case for a broader question: how should AI support human judgment in institutions where safety, service continuity, evidence quality, and operational pressure collide?

Simple stack view

Visibility leads into orchestration, then decision support.

Rapid Report Visibility layer
Meridian Orchestration / workflow layer
Aegis Human-in-the-loop decision support