Ranger Build-A-Bear Hackathon — April 2026

NanuqFi

Strategy Documentation

Protocol: nanuqfi.comGitHub: github.com/nanuqfiProgram: 2QtJ5k...5z5P

540

Tests

27

Instructions

3

Protocols

16.1%

CAGR

01

Strategy Thesis

NanuqFi is a protocol-agnostic, AI-powered yield routing layer for Solana DeFi. Instead of depositing into a single protocol and hoping for the best, NanuqFi dynamically routes USDC across Kamino, Marginfi, and Lulo — choosing the optimal allocation based on real-time rates, risk constraints, and AI-driven market assessment.

The Edge

  • Most yield protocols are single-strategy. NanuqFi routes across multiple protocols simultaneously.
  • The AI keeper scans all protocol rates every cycle, identifies yield gaps, and proposes weight adjustments — validated by algorithmic guardrails before on-chain execution.
  • Architecture survived a live stress test: when Drift Protocol was hacked ($285M, April 1 2026), NanuqFi's protocol-agnostic design enabled full pivot to Kamino/Marginfi/Lulo within hours. Zero user capital lost.

Strategy Type

AI-driven multi-protocol yield optimization
Lending only — no impermanent loss, no leverage

Target APY

10–25%

Depending on risk tier, measured over 90-day rolling windows

02

Operational Mechanics

How capital flows through the system:

1.

User deposits USDC

Into a risk-tiered vault (Moderate or Aggressive) via the on-chain allocator program.

2.

AI Keeper runs every 2 hours

  • Scans live APY rates across all integrated protocols (Kamino, Marginfi, Lulo)
  • AI (Claude) assesses market conditions, protocol health, and rate sustainability
  • Algorithm engine scores each protocol and proposes optimal weight allocation
  • Three-layer validation: AI reasoning → algorithm guardrails → on-chain program enforcement
3.

On-chain program enforces guardrails

Max drawdown, max single-asset exposure, deposit caps, rebalance frequency limits.

4.

User can withdraw

At any time. Two-phase on mainnet (request → redeem after cooling period); instant on devnet.

Three-Layer Validation

AI Layer

Claude assesses market conditions, rate sustainability, and protocol health

Algorithm

Scoring matrix validates proposal against rate, volatility, TVL, and risk thresholds

On-Chain

Anchor program enforces guardrails — drawdown caps, whitelist, weight limits

Protocol Integrations

Kamino Finance

USDC lending via REST API

Live mainnet rates, $1.2B TVL

Marginfi

USDC lending via MarginfiClient SDK

Real on-chain bank data

Lulo

Lending aggregator

Routes across Kamino/MarginFi/Jupiter

Keeper Architecture

  • Algorithm engine with scoring matrix (rate, volatility, TVL, protocol risk)
  • Claude AI layer for regime detection and sustainability assessment
  • Health monitor with cycle tracking, failure recovery, automatic alerts
  • Circuit breaker pattern for API failures (CLOSED → OPEN → HALF_OPEN)
03

Risk Management

On-chain guardrails — enforced by program, not promises:

GuardrailModerate VaultAggressive Vault
Max Drawdown5%10%
Max Single Asset60%80%
Perp ExposureNone (lending only)None (lending only)
Rebalance FrequencyEvery 2 hoursEvery 2 hours
Deposit Cap$10,000$10,000

Drawdown Protection

  • On-chain equity tracking (peakEquity, currentEquity, equity24hAgo)
  • Emergency halt instruction — keeper or admin can freeze all operations
  • Protocol whitelist — only approved protocols can receive allocations

Position Sizing

  • Weight-based allocation (0–100% per protocol, must sum to 100%)
  • No leverage — lending only, zero liquidation risk
  • Protocol diversification enforced by max single-asset cap

Rebalancing Logic

  • AI proposes → algorithm validates → program enforces
  • Rebalance only when rate differential exceeds threshold (avoids churn)
  • Rebalance counter tracked on-chain for audit trail
  • Every rebalance emits an on-chain event with previous/new weights

Stress Test: Drift Protocol HackApril 1, 2026

  • Drift suffered $285M exploit
  • NanuqFi’s protocol-agnostic architecture enabled full pivot within hours
  • allocate_to_drift + recall_from_drift replaced by generic allocate_to_protocol + recall_from_protocol
  • Zero user capital at risk — architecture designed for exactly this scenario
Live Stress Test

$285M Exploit. Zero User Capital Lost.

April 1, 2026

Drift Protocol hacked — $285M exploit

Hours later

Full pivot to Kamino / Marginfi / Lulo — protocol-agnostic architecture enabled instant migration

Result

Architecture proven under real-world adversarial conditions. Generic allocate_to_protocol replaced all Drift-specific code.

This is not a hypothetical scenario. This happened during active development, validating the core architectural thesis: protocol-agnostic design survives protocol failure.

04

Performance

Backtest Results90-day simulation

CAGR

16.1%

vs 5.5% single-protocol baseline

Sharpe Ratio

2.95

> 1.0 is considered good

Max Drawdown

1.89%

worst peak-to-trough

MetricNanuqFi RouterSingle Protocol
Total Return3.75%(90d)1.33%(90d)
Annualized (CAGR)~16.1%~5.5%
Sharpe Ratio2.95—
Sortino Ratio4.86—
Max Drawdown1.89%—

Backtest uses calibrated synthetic data (Ornstein-Uhlenbeck process) fitted to observed protocol rate ranges. Live performance data from devnet keeper available at keeper.nanuqfi.com.

Live Performance

Continuous

Keeper uptime

142+

Decisions logged

98.6%

Success rate

100%

Decisions viewable

05

Technical Architecture

Brief overview — judges will read code:

Monorepo Packages

@nanuqfi/core—Zero-dep interfaces, registry, router, circuit breaker
@nanuqfi/backend-marginfi—Real Marginfi SDK integration
@nanuqfi/backend-kamino—Zero-dep REST API integration
@nanuqfi/backend-lulo—Lulo aggregator integration
@nanuqfi/backtest—Historical simulation engine

27

Anchor instructions

540

Tests across 4 repos

CI/CD

GitHub Actions → GHCR → VPS

540

tests passing

27

on-chain instructions

4

repositories

Green

CI/CD pipeline

NanuqFi

Yield, Routed. Built on Solana.