NexFellow Recognized Builder 2026
Recognized Builder
ζ ω ή

ZOE

A living companion built on a real Drosophila connectome — 139,255 neurons used as a computational reservoir. Not a metaphor. Not an LLM wrapper. Biology that routes.

139K+ neurons (connectome)
11 integrated modules
7-tier falsifiable evaluation
~750 MB reduced footprint
Scroll · Zoe is waking up
NexFellow Recognized Builder 2026
Fellowship Submission · 2026

Project Zoe is my submission
to the NexFellow Fellowship

A self-hosted, bio-inspired agent on a real Drosophila connectome — built with Thought Blend, Thought Resilience, and a 7-tier falsifiable evaluation framework. Recognized by NexFellow as a Recognized Builder 2026 for curiosity, ambition, and commitment to building.

⬇ View the repo — github.com/karun99/zoe 🌐 karun99.github.io ↗ Build mechanism →
Why Zoe is distinct

Three pillars, one living loop

Every part of Zoe is bio-inspired, auditable, and falsifiable. The biology makes the routing decision; the prompt just carries it.

01
🧠

Thought Blend

Focal task + peripheral awareness (screen, history, agent perspectives) merged into one output via an explicit blend rule in the prompt layer. Auditable, not vibes.

02
🛡️

Thought Resilience

The prompt harness isn't static. It re-optimizes and hardens itself across cycles, so output stays coherent under noise, adversarial input, and OOD tasks. Failures get patched, not repeated.

03
🧬

Bio-Inspired Computing

A real connectome substrate with neuromodulation (octopamine / dopamine / serotonin) and Hebbian reinforcement — validated in peer-reviewed research, running locally.

Anatomy

Eleven modules, one presence

Each layer is independently testable. Together they form the Rig — an integrated agent that observes, routes, evolves, and remembers.

System architecture

The Rig — execution flow

One living loop: from connectome features to a persistent, sandboxed companion that returns session after session.

Evaluation

70+ metrics. Zero hand-waving.

A 7-tier falsifiable framework with a Failure Signature Matrix and a Critical Five kill condition — if Zoe fails these, Zoe fails.

↗ rising
Task accuracy
> 80%
Routing accuracy
const / ↓
Cost per neuron
0
Ethical (Vow) violations
< 10%
Generalization gap
"Zoe is a connectome-based reservoir computing system with biologically-inspired control layers, an evolving LoRA adapter, embodied through a virtual avatar, sandboxed in a MicroVM, with companion mode."
— THE HONEST CLAIM · not '80B neurons', not consciousness
NexFellow Fellowship · 2026

Submitted to the NexFellow Fellowship

For demonstrating curiosity, ambition, and commitment to building. This recognition validates the rigor behind Zoe's architecture — not just an idea, but a falsifiable, buildable system.

NexFellow Recognized Builder 2026 Badge

What the Build Mechanism means for Zoe

NexFellow's builder recognition isn't a participation trophy — it's a signal that Zoe ships. Every layer below has been designed to be independently testable, versioned, and deployable. Here's how the build mechanism maps to the Rig:

01
Falsifiable by design
7-tier evaluation with 70+ metrics, Failure Signature Matrix, and Critical Five kill conditions — no hand-waving.
02
Modular & versioned
11 layers, each independently buildable. Bloom, Loom, Drift, Tide — all with their own CI harnesses and JOSS-ready structure.
03
Real substrate, real tools
2 live MCP servers (10 tools), peer-reviewed connectome research, and a Firecracker MicroVM sandbox — not mockups.
04
Resource-honest
~750 MB reduced footprint. No "80B neuron" fantasy. Zoe runs on a laptop, not a data center.
NexFellow Fellowship

Built by Sai Karun Nandipati

Assistant Professor (Data Science & AI) · AI systems engineer · multi-agent architect · NexFellow Recognized Builder 2026

What I bring

  • 2 live MCP servers — 10 tools, stdio, MIT (glama-gateway + s-ai-swarm)
  • JOSS-ready repos with automated validation harnesses
  • A 7-tier, 70+ metric falsifiable evaluation framework
  • Connectome reservoir research lineage (peer-reviewed)
  • NexFellow Recognized Builder 2026 — curiosity, ambition, commitment

Looking for collaborators on

  • Multi-device swarm coordination
  • Voice-to-action pipelines
  • MCP tooling & evaluation design
  • Prototype → JOSS-ready repo hardening
⬇ github.com/karun99/zoe 🌐 karun99.github.io