Profile · AKHAN · Valparaíso, Chile
I build AI you can defend.
Systems grounded in real sources, in domains where a wrong answer has consequences. I study each domain in depth and ship deployed systems that work within its rules.
Biography
From my grandmother, who studied civil engineering, I inherited the phrase that orders how I see the world: mathematics is the language of the world and engineering is the structure of society. I grew up among scientists; I didn't learn the method in a classroom.
I work at the intersection of applied AI, data science and high-consequence domains: I ground my models in real sources so every answer can be verified and nothing is hallucinated. I first proved it in health and aviation, two fields where an error has real consequences. The method, however, does not depend on them: it works wherever there is a hard problem to learn.
“The odd thing is not that I doubt everything but that I translate doubt into architecture.”
Who I am
I communicate
I move easily between teams, clients and specialists from other disciplines (pharmacists, aviation personnel, academics). Working side by side with them is the most direct way to understand a domain precisely.
I lead
I've led teams of 5 to 6 people: I coordinate, define scope, and know when to cut what doesn't add value.
I keep learning
I study every industry I work with down to its full regulation and real practice, and design systems that respect those rules from the start.
I finish
I work a problem until it's done. I don't leave it half-finished or deliver something that barely runs.
I doubt with method
I don't trust a model's output just because it “sounds reasonable”. If it can't cite its source, I don't take its word for it.
I document
I spotted a weakness of mine in documentation and turned it into a strength. I can technically defend any system I build.
Differentiators
I finish.
I take a problem from fuzzy to something someone would actually use. I don't deliver prototypes that barely work: I deliver finished, documented, deployed systems. What matters to me is the distance between the code and someone who uses it, or pays for it.
I build AI you can defend.
Instead of wiring up a model and trusting the output, I ground my assistants in real sources (standards, regulations, domain documentation) so every answer comes with its source and doesn't hallucinate. Where a wrong answer has consequences, that's not an extra: it's the whole problem.
I break problems down and know when to kill one.
I cancelled a project when an equivalent product hit the market, and pivoted. I prefer polished work over merely functional work, but polished means it withstands real use, not that I polish it forever.
From the cockpit to the architecture.
I've flown professional simulators since I was a kid (today the FlightFactor 777), with real procedures, charts, and checklists. From aviation I took a safety culture that software still doesn't have, and I apply it to any system, healthcare or not:
- Procedures are written by accidents. I don't accept a model's output just because it sounds reasonable: traceability is the only thing that separates a system from an opinion.
- Automation never has the final word. In the cockpit no system decides on its own; the captain verifies and can disconnect it. That principle is hardwired into what I build.
- Nothing critical is confirmed with a single source. The cockpit cross-check is the same principle as my retrieval architecture with mandatory citation.
AI + quantum computing.
My path toward graduate school (master's → PhD): how machine learning and quantum computing combine into a system able to propose and evaluate chemical compounds with medical utility: drug discovery.
Honest status: literature review and problem scoping. I study QML and quantum chemistry papers, assess short-term feasibility, and design how to simulate a reduced version. I have academic advising. See reference readings →
- Now · in formulationLiterature reviewQML · quantum chemistry · scoping the problem
- NextToy experimentone narrow task, with a classical baseline
- Graduate schoolMaster's → PhDChina, Japan, or Germany · first preprint in preparation
Trajectory
- Now Ongoing
Degree project in progress with the DGAC
The aeronautical medical fitness assessment (AeroFit), my degree project, is moving toward validation with the aviation authority. In parallel, I'm opening the graduate line in AI + quantum computing.
- 2026
AI in high-consequence domains · PharmIQ
Platform for pharmaceutical chemists with a RAG assistant grounded in regulations, in sales conversations and in use by a real client.
- 2025
Scrum Master · VR chemistry laboratory
Real client (a laboratory director). Unity 6, Meta Quest. Delivered, presented, and praised by the stakeholder; used by ~15 people.
- Professional internship
Tesseract Softwares · software development
Open-source server software: services, Docker integration, and report-grade documentation. ProChile innovation event in the Valparaíso Region.
- Education
Computer Science & Engineering · UNAB
Systems foundations: C++, data structures, and the mathematics that holds everything else up.
Tools
Languages
AI & LLMs
Data & ML
Frameworks & Web
Infra & Data
3D & Hardware
Environment & IDEs
Only tools I have actually worked with, backed by real projects. Emerging ones are marked.
Shall we talk seriously?
Primary role: AI/ML Engineer · Applied Scientist: applied AI and data science. Also: Data Scientist (with modeling) · Research Engineer · ML Engineer · Full-stack with a data or model core. Open to any team with a hard problem (health, aviation, industry, science, critical systems or another domain) where the work calls for understanding the context in depth and delivering a finished system.