Forward-deployed AI engineering · Codex Ambassador · GenAI

Adrián Melic

AI that reaches the repo, the team, and production.

I build AI systems teams can use, evaluate, and maintain. I work from diagnosis and prototype through production and transfer, with a focus on Codex, agents, RAG, voice, and evaluation.

Tell me about the challenge See signals

From technical discovery to operable system, not just a demo
Codex Ambassador and AI educator
Codex, OpenAI API, AWS, agents, RAG, and evaluation
Linked public work and explicit limitations

LinkedIn curso.ai Kaggle

External signals

What you can verify before a call.

Profiles, credentials, and prototypes linked to their public source.

Databricks Certified Generative AI Engineer Associate badge

Verifiable credentials

AWS and Databricks.

The badges link directly to public credentials issued in my name.

Codex · public work

Codex Ambassador.

I build with Codex, OpenAI API, and AWS, and share technical experiments and lessons from my public profile.

See X profile

Kaggle · public profile

Competitions and reproducible work.

The public profile preserves a verifiable history of competition participation.

See Kaggle profile

Services

Three engagements to move from idea to operable system.

I work as a mix of technical lead and builder inside the team's real context: use case, data, repo, risks, prototype, production, and transfer.

Typical 1-2 week scope

FDE diagnosis and production roadmap

I enter the problem with the team, separate what deserves AI from what does not, and leave a defensible roadmap: use case, data, risks, architecture, cost, and delivery sequence.

Technical discovery · roadmap · evaluation · cost · risks

Typical 2-4 week scope

Harness engineering for Codex and agents

I prepare real repositories to work better with Codex and agents: operational documentation, prompts, tests, CI, observability, and review criteria the team can use.

Codex · agent-ready repos · CI · observability · adoption

Typical 2-6 week scope

Prototype to production: RAG, agents, and voice

I build or strengthen a concrete workflow with retrieval, tool calling, voice, human approvals, and traceability. Enough to ship to production without overbuilding.

OpenAI · AWS serverless · Databricks · guardrails

I also run applied workshops, but as part of a real engagement. Training works better when it comes from a repo, use case, or prototype the team needs to operate.

Good fit

This fits when you need someone inside the problem, not another external demo.

You want to move from ChatGPT pilots to measurable production systems.
Your repos, prompts, CI, or documentation are not ready for agents yet.
You need to connect data, tools, human approvals, and traceability.
You need roadmap judgment, but also hands on code, cost, and evaluation.

Track record

Engineering, building, and teaching.

My work combines AI systems, product, and knowledge transfer, with public evidence whenever it is available.

Building with LLMs since the GPT-2 era.
Codex Ambassador and AI educator.
Technical work with OpenAI API, AWS, RAG, and agents.
Public projects linked from this website.

Projects

Product, teams, and repositories that are actual work.

I separate built work from external signals: AI leadership, education product, and public repos when the implementation is ready to show properly.

curso.ai · paused

A practical AI learning system

I founded and designed curso.ai as a hands-on, exercise-based learning experience. The product is currently paused while the training offer shifts toward workshops for teams.

Hands-on learning · product · evaluation

See curso.ai

Student progress in the curso.ai experience

Marketing exercise generated in curso.ai

Talk at Jornada EPA 2025 Aragón about curso.ai

Link's Fairy · public prototype

A browser assistant for reviewing risky pages

A prototype that summarizes page signals, explains risks in plain language, and reuses shared analyses. It supports a decision; it does not provide a security verdict.

Chrome extension · AWS serverless · Mistral

See public repository

Questforge · playable experiment

Codex as a persistent Game Master

A plugin and playable experiment that turns Codex into a Game Master with persistent memory, visible dice rolls, and a visual table. It is an unofficial 5E-compatible tool.

Codex · campaign memory · visible dice · visual table

See Questforge · Public repository

Process

How I work as an AI FDE.

The goal is to leave an operable capability inside the team, not a demo that lives outside its context.

1

Diagnosis in context

What is worth using AI for, what is not, where the bottleneck is, and which data, repos, people, or systems constrain the solution.
2

Harness and working environment

Documentation, prompts, tests, infrastructure, CI, and observability so people and agents can work with less friction.
3

Prototype, evaluation, and production

A workflow, copilot, or agent with cost, latency, errors, guardrails, and traceability visible from day one.
4

Adoption and transfer

Practical training, documentation, and short support so the team can keep moving without depending on an external demo.

Contact

If you want to bring AI to production with your team, let us make it concrete.

I work remotely from Spain. Selected collaborations, technical diagnosis, implementation sprints, harness engineering, and practical training for teams.

If you send context, the affected repo or process, the team, and the timing, I can respond with more precision.

Send me an email

Include the context, the affected process or repo, the team, and the timing. That will help me respond with a useful first view.

info@adrianmelic.com Contact me on LinkedIn