Aldana Rodriguez
Brain dumping
Welcome to my digital mind. Not sure yet how this will look like in the future.
Projects
AnkiQuestSep 2026A gamified fork of Anki, aimed at increasing usage and retention. This will soon go public for everyone to use, I'll link it here as soon as that happens.
Pet Health MemoryJul 2026A health memory and longevity tracker for pet parents.
Script Kiddie UpliftICLR 2026How mid-2025 LLM-generated code uplifts novice attackers.
Keeping myself honest
Writing
Writing
Nothing here yet.
Projects
Things I've built and studied
Software · Sep 2026
AnkiQuest
A gamified fork of Anki, aimed at increasing usage and retention. This will soon go public for everyone to use, I'll link it here as soon as that happens.
- Stack
- Kotlin, Android
- Built with
- AI coding assistants
What I built.
- Native AnkiQuest screens, a notification inbox and a review heatmap.
- Account-scoped sign-in and safeguards on in-app web navigation.
- Regression tests for account changes, redirects and credential handling.
- Fixes for app lifecycle, build and lint issues in a large inherited codebase.
I intend on writing a post going into more detail soon.
Software · Jul 2026
Pet Health Memory
A health memory and longevity tracker for pet parents.
- Stack
- Next.js, TypeScript, Supabase
- Built with
- Codex (GPT-5.6)
- Works on
- Web, installable on phones
What it does.
- Keeps each pet's medications, reminders, vet contacts, expenses, microchip and insurance details in one place.
- Turns that history into summaries and questions to discuss with your vet.
- Exports a Vet Packet to share or print before an appointment.
- Stores profile photos, symptom photos and vet documents privately.
- Offers an optional AI summary of what changed recently.
Research · ICLR 2026 workshop
Script Kiddie Uplift
We study how mid-2025 LLM-generated code uplifts novice attackers, and whether models assist harmful goals when requests are decomposed into benign substeps.
- Venue
- ICLR 2026, Agents in the Wild workshop
- Authors
- Che, Poveda, Rodriguez, He, Zhang, Wang, Sehwag
- My role
- Research operations, data gathering, quality assurance
- Step-by-step interactions rarely trigger refusal, even when the same models recognize safety concerns in separate evaluation.
- Code-assisted novices progressed nearly one additional step further on average than an internet-only baseline.
- Weaker performers benefited disproportionately, closing on average 47% of the gap to top performers.
If you think I could help with your research, do let me know!
Earlier · machine learning notebooks
Learning experiments
- BERT text classification: fine-tuning BERT with TensorFlow.
- English to French translator: a sequence model built with LSTMs.
- Life expectancy predictions: linear regression on country data.
Now · updated October 2026
What I'm focused on
A short - non exhaustive - list of current priorities. If it isn't here and you think it should, please reach out to me!
- Leading LinkedIn's Agentic Pod, which builds reinforcement-learning environments, gyms and other agentic data.
- Building a network of mental healthcare professionals aimed at people that work in AI Safety.
- Finishing a B.S. in Computer Science at the University of the People, due January 2027.
- Getting this site going
- Thinking about how I can best help AI go well.
About
About me
I'm Aldana, Argentine born and raised but now based in Berlin. I'm the head of the Agentic Pod at LinkedIn AI, where I lead data programs for reinforcement-learning environments and agentic workflows. Before that I ran human-data programs for frontier AI labs at Scale AI, including a cybersecurity research project with more than 50 PhD candidates.
What I care about most is reducing (and ultimately ending) suffering of all sentient creatures. This includes animal welfare, but also reducing the risks that come with AGI and ASI.
I've taken the Giving What We Can 10% Pledge: every month I donate 10% of my salary through Effektiv Spenden.
Contact
aldanabrz@gmail.com
Mistakes
Things I got wrong
Each entry gives the date, the original claim, what's true instead, and who caught it.
Nothing logged yet.
The site is new. If you spot an error in anything here, tell me and it will appear on this page with credit.
Changelog
Site changelog
New pages and revisions, newest first.