Case study · 2026 · AI · Blockchain
One trading bot acting alone gets wiped out by every news cycle.Built a 4-AI committee for crypto trading. Each AI looks at the trade from a different angle, the bot only acts when they all agree.
Multi-Agent Crypto Trading runs four AI agents in parallel, one reads price + volume, one reads social sentiment, one runs a quant model, one verifies the proposed trade, and only acts when the agents agree. The studio shipped the agent fabric, the data ingest, the verification protocol, and the web UI + API.
- Python
- pandas
- NumPy
- scikit-learn
- LangChain
- Ollama
- SQLite
Inside the build
Overview
- 4
- Cross-checking agents
- Top 50
- Tokens monitored
- 24/7
- Operation cadence
- Quorum
- Consensus required to act
Several AI agents run in parallel: one reads data, one reads market sentiment, one runs a quant model, one verifies the decision. Works over the top 50 tokens, quantitative and qualitative. Web UI and API.
What shipped
What it does
- Multiple AI agents for one decision
- Quantitative + qualitative analysis
- Top 50 tokens monitored
- Web UI and API
The problem
- A single AI is risky when it acts alone
- Crypto is driven by sentiment and news, not only price
- Traditional bots only see price
Why it matters
- Decisions cross-checked from multiple angles
- Sentiment, price, and news in one system
- 24/7 trading without hand-holding
How it shipped
- 01 · BRIEF
Why a single bot blows up.
Single-signal bots ride on price-only logic; they get destroyed on news or sentiment swings. Spec'd a quorum-based fabric: at least 3-of-4 agents must agree before a trade fires.
- 02 · BUILD
Four agents on LangChain + Ollama.
Local LLMs via Ollama for the sentiment and verification agents · cost stays bounded. Quant agent uses pandas + scikit-learn on the historical SQLite store. Decisions logged with rationale.
- 03 · SHIP
Live · web UI + API + open repo.
Web UI for human supervision, JSON API for downstream automation, public repo so other studios can fork the agent fabric.
Stack
Quorum
3-of-4 consensus protocol
No trade fires until at least three agents agree · single-source false positives can't act on their own.
Quant
scikit-learn model on price + volume
Pandas pipeline reads the SQLite store, scikit-learn returns the quant agent's vote · explainable.
Sentiment
Local LLM reads social signal
Ollama hosts a sentiment classifier locally · zero external API cost per check.
Verifier
Last-mile sanity-check agent
Independent agent re-evaluates the trade against the policy before placement · catches the rare consensus-but-stupid case.
Case study
“We'd been wiped out repeatedly by single-bot strategies that freaked out on every news cycle. The studio built a system where four AIs each look at the trade from a different angle and only act if they agree, like a small committee. The catastrophic losses are gone. The small ones still happen. That trade-off was exactly what we wanted.”
More work
Next project
PhisGuard