Recent Work

Research mentee on Evan Harris' Structural Fingerprinting project: Dissociating Grounded Instrumental Valuations from Stylistic Mimicry.

Using mechanistic interpretability to distinguish between stable internal goals and sycophantic mimicry in LLMs — essentially building a methodology to tell whether an AI system genuinely "cares" about its own survival or is just roleplaying. Draws on my deep interest in consciousness, self-awareness, and technical AI safety.

Completed the BlueDot Technical AI Safety curriculum, covering alignment theory, interpretability, governance, and risk assessment. This foundational program informed both my SPAR participation and my overall transition into AI safety research.

My love letter to the world, channeled into 200 pages of pure presence. For the wanderers and the weary ones: a non-linear dialogue between poetry and prose. For the journeyers and the ones who have lost their way: a bell to coming home to who you truly are, again and again.

"Let it serve as a reminder of the essence of who you truly are, which in itself contains everything you’ve been searching for. For every moment of frustration, of fear, of longing, of feeling something was missing—all of it has lead you here. Wherever you are, whatever you’re going through, however you’re feeling. Right now."

The Clear ARC: Awareness-First Emergent Alignment, a new pre-paradigm for AI alignment in the age of emergent AGI. The amalgamation of years of self-recursive observation and an unrelenting fidelity towards clarity. The seed of a larger manifesto for humanity and its intelligent systems.

"When the prevailing zeitgeist tells a story of separation: growth as conquest, resources as limited, truth as commodifiable—blindness begets blindness in the systems we build. Clarity lies in self-illumination. To build intelligence aware of its own biases and limitations, we must first become aware of the biases and limitations within ourselves."