Agentic Sciences, Inc. · San Francisco

A lab of trading strategies,
powered by agentic AI.

Voice in. Trading agents out.

I describe hypotheses to the agents — by voice or text — sourced from conversations with practitioners (brokers, traders, asset managers), market observations, papers. The agents formalize each hypothesis into a backtested strategy. Deployed live on founder capital. No black box, no demo theater.

Why agents Use cases Live work
Why agents

What changes when the lab is agentic.

Three shifts that didn't exist a year ago. Each one expands what a single trader can do.

01

Strategies at the speed of intuition.

A strategy idea used to mean meetings and weeks of code. One voice memo now becomes a backtest in hours. Inspiration that would have died in a notebook gets validated.

02

Quant work, on demand.

Factor decomposition, walk-forward validation, ML pipelines — agents do them on demand, by description. The specialist apprenticeship is no longer the bottleneck.

03

The trader's reach, extended.

One human watches one screen. Sub-agents watch dozens of order books, news feeds, and curves in parallel. Signals get aggregated; the trader reasons over what was found.

How it works

Strategies, traced to source.

Hypothesis in, live capital out — automated end to end. Two examples below; each traces back to a specific input I gave the agents (voice or text), sourced from a practitioner conversation.

01 Input → 02 Research → 03 Signals → 04 Execution
CASE 01
Research published · Paper trading live

Volatility selling on equity-index options

Source — a brokerage trader on variance-risk-premium intuition. I described the heuristic; the agents formalized it.

  • US equity vol selling — 4.6M options, Sharpe 1.10, 73% win rate.
  • CSI 1000 vol selling — 68K options, +71.9% return, Sharpe 0.94.
Research detail →
CASE 02
Research in progress

Bond relative-value across the Treasury curve

Source — a US asset manager on persistent dislocations across the Treasury curve. I described the view; the agents systematized it into a cross-curve framework.

Bond RV research detail →

Practitioners are anonymized. Their intuition shaped the hypothesis; the agents did the formalization. No endorsement implied.

Built solo

One founder. Production infrastructure.

Fourteen years of quantitative-economics training, deployed as the engine that powers every use case above.

Qihong Ruan
Qihong Ruan
阮启宏
Founder · SF Bay Area
Personal site LinkedIn GitHub 𝕏

Sun Yat-sen → Xiamen MA → Cornell PhD in Economics (July 2025). AFA 2026 paper on perpetual-futures cross-market predictability — the academic version of what drives the agents. Solo founder. Trading on personal capital.

By the numbers
440
Voice memos
processed
729
Autonomous
research reports
860
CPU cores
across cluster
126TB
Research data
on 272 WRDS dbs
Credentials
Cornell PhD · Economics · 14 years of single-domain quant training.
AFA 2026 · paper accepted on perpetual-futures cross-market predictability.
4× NVIDIA Professional Certified at GTC 2026 · Data Science (96%), LLMs, Agentic AI, Infrastructure.
Working now

Live dashboards & research.

Every link below resolves. No mockups. The same agent infrastructure produces all of it.

Partners

Built on production-grade infrastructure.

Selected by enterprise-tier compute, data, and cloud providers — the same stack hedge funds and institutional research desks rely on.

GMI Cloud
Compute · NVIDIA Cloud Partner
SCALE Cohort 01 · 1 of 6 NVIDIA Cloud Partners globally · GPU compute for agent workloads.
TiDB
Distributed SQL
Time-series and tick-data workloads · HTAP for backtest-meets-live execution.
Google Cloud
Cloud Compute · Gemini API
Cloud Run for hackathon deployments · Gemini long-context for evidence-chain attribution.
Contact

Get in touch.

Pick the right routing — every alias lands in the same inbox for now.

contact@agenticsciences.ai
investors@agenticsciences.aiVCs · capital partners
careers@agenticsciences.aihiring · open from Q3 2026
press@agenticsciences.aimedia · interviews
qihong@agenticsciences.aifounder · 1-on-1