The problem Mochifi solves

🛡️ The Problem It Solves

Demo: https://share.descript.com/view/pnsbObRikUV

The Challenge: Information Chaos & Adversarial Markets

Traders and participants in futarchic decision markets (such as Umia, MetaDAO, and Polymarket) operate in a high-stakes, high-noise environment. Currently, making an informed decision requires manually tracking scattered signals across Telegram, X (Twitter), Discord, GitHub activity, and on-chain movements.

This fragmented landscape leads to critical failures:

• High Cognitive Load: Participants spend hours cross-referencing data, leading to slow and inefficient decision-making.
• Vulnerability to Manipulation: Markets are often swayed by coordinated misinformation and "pump-and-dump" schemes disguised as community consensus.
• Inefficient Capital Allocation: Poor information quality results in mispriced risks in high-stakes governance and prediction markets.

The Solution: Umia Analyst Agent

The Umia Analyst Agent is an autonomous AI research analyst designed to turn chaotic data into actionable, auditable intelligence. It operates as a continuous intelligence layer that:

• Smart Ingestion & Weighting: Automatically classifies signals from multiple sources using a credibility-weighted model (Verifiable / Partial / Manipulable).
• Adversarial Synthesis: Generates structured Bull and Bear cases with direct citations, ensuring every claim is backed by evidence.
• Anti-Manipulation Shield: Actively flags coordinated manipulation attempts by cross-referencing social hype with on-chain reality.
• Skin in the Game: Beyond mere reporting, the agent publishes structured trading memos, takes its own positions, and maintains a public, transparent PnL track record.

🚀 Key Benefits

The Umia Analyst Agent doesn't just provide data—it provides the "Signal" necessary to navigate the "Noise" of decentralized governance.

Challenges we ran into

Challenges WE Ran Into

Building the Umia Analyst Agent presented several unique technical and conceptual hurdles. Our goal was to move beyond simple "AI wrappers" and build a resilient, institutional-grade engineering system.

1. Adversarial Information Filtering

The biggest challenge was creating a system that doesn't just aggregate data but actively resists manipulation. In futarchic markets, we found that social signals (Telegram/X) are often heavily manipulated by proposal authors to create "fake consensus."

• The Solution: We implemented a Source Classification Layer. We assigned deterministic weights to data: verifiable sources (on-chain transactions, GitHub commits) are weighted at 1.0, while manipulable social channels are strictly capped at 0.3.

2. Ensuring Agentic Reliability

Standard LLM prompts often suffer from "hallucinations" or "groupthink." In a financial context, we couldn't risk the agent making biased or "hallucinated" trading recommendations.

• The Solution: We utilized a LangGraph-based multi-node architecture. By separating the pipeline into distinct stages—Source Classification, Synthesis, and Position-taking—and adding an autonomous Self-Reviewer node (the "Advocatus Diaboli"), we forced the agent to argue against its own conclusions before finalizing a memo.

3. Domain Shift & Architecture Repurposing

We started with a robust pipeline designed for corporate credit underwriting and had to transform it into a DeFi intelligence tool within 48 hours. Mapping "banking risk" to "governance risk" required a total overhaul of the business logic.

• The Solution: We maintained the underlying "plumbing" (deterministic logic gates, audit trails, and versioned skills) while completely swapping the domain layers. This allowed us to preserve institutional-grade reliability while moving at hackathon speed.

4. Data Ingestion Latency

Scraping decentralized forums and GitHub activity in real-time is computationally expensive and prone to rate limits.

• The Solution: We developed a dual-mode ingestion system. For the demo, we use a high-fidelity mock-market layer based on historical MetaDAO proposals, while the production environment is designed to utilize Apify actors for fresh signal ingestion.

5. Ecosystem Early-Stage Adaptation (The MetaDAO Bridge)

Since Umia’s native decision markets are currently in their earliest stages, we faced a "cold start" problem regarding live data for testing our agent's predictive accuracy.

• The Solution: We chose to stress-test our agent on MetaDAO (the pioneer of futarchy). This is a strategic advantage: by proving our agent can navigate the most adversarial and "noisy" established market in the space, we ensure it will be the most robust and battle-tested analyst the moment Umia’s mainnet markets go live. We aren't just building for a sandbox; we are building for the open ocean.