Meet the fintel-optimized agent

We've built an in-house agent through financial AI evaluations. Deployed since April this year, it systematically covers the Dow Jones universe to generate active investment calls. No human in the loop.

This live strategy is a demonstration of what evaluations unlock, and our skin in the game.

Gross return since deployment

$135,394.88NAV (USD)
-5%0%5%10%15%20%04/2405/1406/0406/2507/1608/0508/2509/1509/30
In-house agent (gross)Dow Jones (price-weighted)Dow Jones (market-cap-weighted)

Alpha vs price-weighted

-2000+200+400+600+80004/2405/1406/0406/2507/1608/0508/2509/1509/30
Alpha (bps)

Past performance is not indicative of future results.

Evaluated and optimized for performance

Through fintel, we evaluated at scale across the agentic components — model, harness, data, prompt — on the investment KPIs we care about. A model is only chosen because it works best for the specific investment strategy and harness.

Agent eval: Backtest performance

Choosing the model

0.901.261.621.982.3420222023202420252026
DJIAhigh intelligence + hallucinationlow intelligence + hallucinationbalanced traitshigh intelligence + low hallucination

Agent eval: Backtest performance

Customizing the harness

0.901.261.621.982.3420222023202420252026
DJIAtool-calling ReActtool-calling Langgraphdata-feeding Langgraphhybrid

Agent eval: Drawdowns

Evaluating additivity of a dataset or prompt

-0.05-0.04-0.03-0.010.002026-012026-022026-032026-042026-052026-062026-072026-08
original strategylow-volatility iteration

Agent eval: Trading costs

Finding the right rebalancing cadence

0.07.214.521.7Ann turnover (×)10.3Biweekly18.4Weekly0.000.320.640.96Ann trading cost (%)0.48Biweekly0.81Weekly
BiweeklyWeekly

Evaluated and controlled for AI-specific risks

When AI is generating alpha, AI-specific risks = investment risks. Through fintel, we evaluated agent stochasticity and hallucinations to impose the right controls.

Agent eval: Stochasticity across identical repeats

Same agent, same date, same ticker (JPM) — different scores

-0.38-0.150.080.300.5320222023202420252026
Repeat 1Repeat 2Repeat 3

Agent eval: Backtest performance

Systematic vs unconstrained trading

0.890.961.041.111.192026-012026-022026-032026-042026-052026-062026-072026-08
MVOUnconstrained (agent trade like PM)

Post-deployment outputs and evals

fintel houses our pre- and post-deployment agents under the same roof so we can continuously evaluate and fine-tune our agents. Raw outputs and evals are showcased below.

2026-09-2818 holdings
AMGNAMZNAXPCRMCSCODISGOOGLGSHDJPMMCDMSFTNVDASHWTRVUNHVWMT

Each row is the in-house agent's score and rationale for a DJIA constituent on the decision date. Tap any row to expand the rationale and key factors.

What we offer

fintel evals

  • Eval database

    • –

      Access to our proprietary eval database across common models, harnesses, and datasets.

    • –

      Go beyond headline metrics: agent outputs, behaviors, biases, and risks.

  • Proprietary research

    • –

      Access to commentaries and guides from our proprietary evals and in-house alpha strategy.

Customized evals

  • For AI investors

    • –

      Evals tailored to your agentic specification and vision — choose the model, harness, data, and strategy.

  • For AI product providers

    • –

      Customized evals on your AI product — quantifying the additivity of your data, model, or harness to AI agent alpha.

Generate alpha

  • In-house agents

    • –

      Access to our in-house alpha agents, continuously evolved through our proprietary evals.

  • Investment signals

    • –

      Leverage the alpha signals generated by our in-house agents, each developed and tested through our institutional-grade systematic research pipeline.

Credibility

Evals done by industry practitioners and pioneers with skin in the game.

Scalability

Our proprietary eval platform, built for any agent and any financial task.

Read more about us

Scale AI alpha with us today

founders@fintel.capital