← Back to Home
Case Studies

Systems in production.

Not prototypes or demos -- real autonomous systems managing capital and processing data around the clock.

Autonomous Trading

AI Agents Managing Real Capital 24/7

A multi-strategy autonomous trading system that executes, manages risk, and improves itself -- with zero human intervention.

CHALLENGE

Manual crypto trading is emotionally driven, inconsistent, and can't operate around the clock. A human trader makes impulsive decisions under pressure, misses opportunities during sleep, and can't simultaneously monitor dozens of signals across multiple strategies. The result: inconsistent returns, blown stop-losses from panic, and missed 3 AM setups.

SOLUTION

Built a multi-strategy autonomous trading system with AI-powered risk management. Multiple independent strategy agents run concurrently, each specializing in a different market pattern -- momentum, whale order flow tracking, regime detection, and more. A central risk engine enforces position limits, drawdown protection, and capital allocation while a nightly learning daemon (Ralph) analyzes all trades and auto-tunes parameters.

KEY FEATURES

Multiple concurrent trading strategies (momentum, whale-tracking, regime detection)
ATR-based dynamic stop losses that adapt to market volatility
Circuit breakers for drawdown protection (3 consecutive losses = 4h cooldown)
Real-time Discord notifications for every trade, alert, and risk event
Automated nightly learning loops (Ralph daemon) for continuous parameter tuning
Tiered LLM routing to minimize API costs while maintaining intelligence

RESULTS

24/7
Autonomous Execution
66%
Best Strategy Win Rate
<1s
Execution Latency
0
Emotional Decisions

TECH STACK

Python Hyperliquid API VPS Deployment systemd Services Discord Webhooks
Knowledge Pipeline

Turning Any Content Into a Searchable Knowledge Graph

An AI-powered knowledge engine that transcribes, extracts entities, maps relationships, and creates a fully searchable knowledge graph.

CHALLENGE

Valuable knowledge locked in videos, podcasts, and long-form content is impossible to search or connect. A single 2-hour podcast might contain dozens of actionable insights, references to people and companies, and connections to other topics -- but none of it is indexable. Traditional note-taking captures fragments; the relationships between ideas are lost entirely.

SOLUTION

Built an AI-powered knowledge engine that ingests content from any format, automatically transcribes audio/video, extracts named entities using NLP, maps relationships between entities across all ingested documents, and surfaces everything through full-text search and an interactive knowledge graph. Available as both a desktop app (Tauri) and web SaaS.

KEY FEATURES

Multi-format ingestion (YouTube, podcasts, PDFs, raw text)
Entity extraction with NLP for people, organizations, and concepts
Automatic relationship mapping across all ingested documents
Full-text search with TF-IDF ranking via Tantivy engine
Interactive knowledge graph visualization for exploring connections
Timeline view for chronological exploration of extracted events

RESULTS

10x
Faster Knowledge Retrieval
Auto
Entity Linking Across Docs
Visual
Knowledge Graph Exploration

TECH STACK

Rust (7 crates) SvelteKit Tantivy Search SQLite Supabase Tauri (Desktop) Web SaaS

Ready to build something autonomous?

These are the kinds of systems we build. If you have a problem that needs an AI agent -- not just an AI feature -- let's talk.

Start a Project