Helios Labs · 2025

Realtime trading intelligence

A sub‑200ms trading copilot that ingests order books, news and sentiment to surface high‑signal opportunities for a mid‑market hedge fund.

Tag
AI · Web
Duration
14 weeks
Role
Design · Engineering · ML Ops
Realtime trading intelligence
Decision latency
−72%
Alpha (90d)
+18%
Eval coverage
97%
◇ The problem

Where they were stuck.

Helios analysts were drowning in 6 separate dashboards and lost an average of 90 seconds per decision. Latency was killing alpha.

◇ Our approach

How we shipped it.

  1. 01

    Replaced the multi-dashboard sprawl with a single, keyboard-first command surface.

  2. 02

    Built a streaming feature store on top of ClickHouse + Redis with sub-50ms p95.

  3. 03

    Trained a domain-tuned LLM judge to rank news/sentiment events against open positions.

  4. 04

    Wrapped every model decision with a full eval harness and human-in-the-loop overrides.

◇ Stack

What it's made of.

Next.jsReactRustClickHouseRedisOpenAIVercel
Next case study

Zero‑trust security mesh

Sentinel

Have a problem this hard?

Start a project →