Neeraj Venna
Type a command to browse the work.
I’m interested in world models, agentic memory, and decision-making under uncertainty. My project, BeliefSpec, looks at whether predicting future observations helps agents remember useful information.
An agent sees a clue, loses sight of it, and must remember it to choose the correct path.
- Ready
- Observation
- Memory
- Decision
- Outcome
Agent State
Start the trial to see a clue.
- Current observation
- Corridor entrance
- Remembered clue
- Not seen yet
- Available actions
- Forward
- Selected action
- —
- Outcome
- Pending
Illustrative demo · rule-based decisions. Changing memory or switching the clue restarts the trial.
PREDICTIVE MEMORY / 2026
BeliefSpec.
Does predicting the next observation teach a system to remember what matters later?
A controlled study of whether next-observation prediction helps recurrent memory retain information for a later decision.
View project
01 / RESEARCHBeliefSpec
A controlled study of whether next-observation prediction helps recurrent memory retain information for a later decision.
Projects
All projects
Basis Set Ventures
AI Fellow
Evaluated agentic memory through temporal recall, conflicting information, irrelevant context, and paraphrased questions.
Salk Institute
AI Research Intern
Built behavioral analysis and pose quality tools for ALS research.
Scripps Research
Research Assistant
Supported brainstem interoception research through two-photon imaging and neuronal-region analysis.
Network Contagion Research Institute
Data Engineer Intern
Built data tools and application components for incident-report analysis and geospatial analysis.