Kavana Venkatesh

Kavana Venkatesh

CS PhD Student · Agentic AI · Self-Evolving Agents

Homepage V3 — Layout and editorial copy

Homepage V3 — Layout and editorial copy

Structure (desktop)

+---------------------------------------------------------------+
|  Kavana Venkatesh                Home  Research  Publications... |
+---------------------------------------------------------------+
|  Portrait / pastel-backed print  | Hello, I'm Kavana.           |
|  CS PhD Student · Virginia Tech  | PhD / affiliation            |
|  Scholar · LinkedIn · GitHub     | MY RESEARCH · IN BRIEF        |
|  Email · CV                     | technical summary            |
|                                  | personal line + jump links  |
+---------------------------------------------------------------+
|  01  RESEARCH NOTES        four interactive notebook tabs       |
|  02  SELECTED PAPERS       four formal paper records; GIF slots |
|  03  NOTICEBOARD          recent notes + filter + Old News      |
|  04  IN THE MARGINS       a small invitation to personal page  |
+---------------------------------------------------------------+

On mobile, portrait and profile links stack above the introduction; the notebook tabs use a 2x2 index. The full publication records and dated news remain below.

Academic identity

  • Multi-Agent Safety & Coordination
  • Self-Evolving Agents
  • Emergent Behaviors of Multi-Agent Systems
  • Agent Harnesses & Evaluation

These represent four connected aspects of agentic AI, not mutually exclusive research categories.

Hero copy

Hello, I’m Kavana.

I’m a Computer Science PhD student at Virginia Tech, advised by Dr. Jiaming Cui and affiliated with the Sanghani Center. I’m also a PRISM AI Safety Research Fellow.

MY RESEARCH · IN BRIEF

I study multi-agent safety and coordination, self-evolving agents, and emergent behaviors in LLM-based multi-agent systems. My work investigates how agents communicate, delegate, and develop collective patterns; how failures propagate through memory, tools, and communication; and how adaptive agent harnesses can help us monitor, evaluate, and safeguard increasingly autonomous systems.

More broadly, I’m interested in the relationship between agent autonomy, collective behavior, and system-level reliability—both empirically and through methods that make these systems more interpretable and dependable.

Outside the lab, I’m also a writer and editor. Poetry, philosophy, photography, and bringing people together to learn are all part of this little world.

Research notebook

  1. Safety & coordination — cascading attacks, causal attribution, security and reliability across memory, tools and communication.
  2. Self-evolving agents — capability changes in skills, workflows, tools, and memory; safety adaptation as systems evolve.
  3. Emergent MAS behavior — heavy-tailed interactions, empirical coordination laws, emergent influence, scaling bottlenecks.
  4. Harnesses & evaluation — orchestration, monitoring, controlled interventions, and assessments of reliability, calibration, and uncertainty.

Selected papers

The homepage contains four featured 2026 papers with paper/project links and empty future-figure GIF spaces. Do not replace the placeholders with invented scientific diagrams. A complete publication archive remains a separate page.

News

The existing retro notice board is retained, with All / Research / Life filters and the Old News drawer. The archive keeps the earlier dates and links from your original homepage. Confirm the conflicting early iLink Digital dates before adding that item back.

Visual principles

  • One portrait with a single mint backing sheet, not a pile of cards.
  • Vintage typewriter labels are accents; technical paragraphs use readable proportional type.
  • Scholar, LinkedIn, GitHub, Email, and CV links are immediately beneath the portrait.
  • Research substance leads. The literary personality is visible but doesn’t obscure scholarship.
  • Motion is responsive to user interaction, not autoplay marketing animation.