Try the Konami code:
↑↓↑↓←→←→BA
Hi, I'm Naavya, an independent researcher in AI systems and cognition.
Current direction: intrinsic motivation as a structural answer to reward hacking — turning abstract psychological concepts into testable criteria for alignment.
I'm a recent graduate from the University of Illinois at Urbana-Champaign, where I studied computer science, philosophy, and psychology.
These are my most promising projects. Click to learn more.
Introduces MIRA (Motivated Intrinsically through Relational Anchoring), arguing reward hacking is a failure of motivation, not specification. Drawing on Hegel, Vygotsky, and Self-Determination Theory, it develops agents whose goals hold up once external reward is removed, with operational, testable criteria for intrinsic motivation.
Introduces PMS 2.0 (Preprocessing Metacognitive System), a system-agnostic upstream routing layer that decides when to engage, escalate, or refuse tasks before downstream reasoning. Demonstrates improved conditional accuracy, lower compute load, and interpretable governance of computational effort.
Exploratory frameworks and prototypes built to think through problems in systems, learning, and representation. These projects are not production tools — they’re attempts to make abstract ideas concrete, test assumptions, and see where models break under pressure.
The question I get every time my background comes up, from cab drivers to conference peers. It seems to be a burning one, so here's how I think about it.
Every AI system is shaped around human intelligence: trained on what we write, evaluated against what we judge, and aligned to what we value. Whatever we build, a human mind is the reference point.
Philosophy spent millennia asking what minds, knowledge, and values are. Psychology turned those questions into experiments and measurable constructs, the longest-running study of intelligence we have.
AI's hardest open problems (reward hacking, sycophancy, deception, values that don't generalise) are old questions in a new substrate. Psychology already has decades of theory and failure data on how motivation and values form, and how they break.
So philosophy and psychology aren't a detour on the way to frontier AI, AGI, or recursive self-improvement. They're a head start. The work is translating their concepts into criteria we can actually test.
For collaboration, feedback (rebuttals especially welcome), or a good conversation about alignment or anything else I'm curious about, I'd love to hear from you.
