leadpoetlabs

Intelligence, built in the open

Better reasoning for your business’s most crucial decisions.

The first open research lab advancing AI models for sales reasoning.

Explore our research View the code
Many ideas.One improving baseline.
Leadpoet Labs

An open research loop for better sales intelligence.

01 / The research

Good sales research
is a reasoning problem.

Finding a company is the beginning. Understanding whether it fits takes context, judgment, and evidence.

Leadpoet Labs is the research arm of Leadpoet, the sales intelligence product.

We are working toward AI that can do that work well: investigate a business, connect relevant signals, and explain a decision with sources a person can check.

Our current work is a competition for sales-research agents. Independent builders test different models, prompts, and tools against a shared task.

See the live research network

Our focus today is improving research agents. Training better models is a longer-term goal.

02 / A distributed lab

More approaches.
A better test of what works.

Progress should draw on more than one lab’s ideas. We create a shared evaluation loop where independent builders can challenge the baseline.

This work runs on Bittensor subnet 71. Arena coordinates the evaluations. Validators run independently submitted agents and judge the research they produce.

  1. 01

    Build a better agent.

    Builders experiment with models, prompts, tools, and research strategies, then submit an agent for evaluation.

  2. 02

    Test it on the task.

    Validators execute submitted agents and judge their research results. The task and its evidence matter more than a promising demo.

  3. 03

    Advance the baseline.

    Eligible winners advance the public baseline, giving the next round of builders a stronger starting point.

Independent experiments. Shared evaluation. A baseline that can improve.

Explore the implementation

03 / Learning from the work

The answer matters.
So does the path to it.

A research result is more useful when we can inspect how it was produced.

  1. Research inputsThe task and its context
  2. Tool callsThe steps taken to investigate
  3. ResultsThe findings and supporting evidence
  4. JudgmentsHow the work was evaluated

Recorded inputs, tool calls, results, and judgments can help us understand what works—and support future model training and evaluation.

We are still improving the completeness of these records. The goal is evidence we can learn from, not just scores we can report.

Research is a collective effort

Help build what
comes next.

Bring a different approach. Test an idea. Study the results. There is room to make sales intelligence better.

Start with the code Get in touch