Hey, I'm Adithya!
I work on multi-agent simulations and autonomous research systems at Exaforge Labs.
I'm broadly curious about how the world works and tend to go deep on whatever catches my attention. Right now, my focus is on what current AI systems structurally can't do yet, especially around taste, intuition, and the kind of creativity that reframes a problem instead of just solving it.
Featured Projects
Research
All researchSearch Discipline for Long-Horizon Research Agents
A search governance protocol for long-horizon research agents that audits candidate effects across disaggregated regimes and uses an external monitor to govern selection and stopping.
Video Anomaly Detection using Factorized Self-Attention Transformer
A weakly supervised audio-visual anomaly detection pipeline that combines factorized spatial and temporal attention, spectrogram transformers, bidirectional cross-modal attention, and robust temporal feature magnitude learning.
A Transformer Approach for Weakly Supervised Abnormal Event Detection
A weakly supervised video anomaly detection pipeline that extracts hierarchical shifted-window transformer features from frame sequences and trains a classifier using video-level labels.
Essays
Essay indexBeyond the Given Frame
Why the hardest form of creative thought may be recognizing when a problem cannot be solved inside the assumptions that made it legible.
The Intelligence to Know What Matters
Why generating possibilities is different from recognizing a direction worth pursuing, and why judgment depends on memory, consequence, and time.
Can Synthetic Populations Predict Real Behavior?
A benchmark of where synthetic populations recover real-world behavior, where they miss, and how the evaluation avoids hindsight.
Building a Population Simulation Engine
A technical account of the dependencies, networks, state, and timesteps behind a synthetic population engine.