A new peer-reviewed study ran 61,814 startups from Freigeist Capital’s pipeline through an agentic LLM and compared the results to Freigeist’s own analysts.

A human analyst needed around 2 hours for a thesis-driven screen. The LLM agent did the same job in about 13 seconds, a 537× speed-up for the same class of screening.

On screening quality, the agent stayed close to human performance on a standard clustering metric:

• Human analysts: 0.37 Silhouette

• LLM agent: 0.35 Silhouette

On the Calinski–Harabasz index, the LLM scored about 70% higher, which signals tighter, better-separated clusters across more than 61k ventures.

In follow-up data, startups surfaced by the LLM were more likely to survive and more likely to raise follow-on funding than the baseline group, including some that human analysts did not highlight. Ventures selected by both the agent and the analysts showed the strongest survival and funding rates.

Taken together, the study shows that an LLM-driven screening layer can run on real VC dealflow at scale, stay within the performance band of experienced analysts, and surface companies with stronger basic outcomes.

This is the layer ADIN has treated as core from the beginning.

The Freigeist study shows that an LLM-based screening layer can:

• run thesis-driven screens about 537× faster than a human analyst

• stay in the same quality range on Silhouette score

• deliver significantly higher cluster separation on Calinski–Harabasz

• surface startups with higher survival and follow-on funding rates, with the strongest signal where model and human selections overlap

ADIN is built for that hybrid.

• Scouts broaden and diversify the pipeline

• LLM-native infrastructure structures and screens that pipeline at scale

• Investors concentrate on selection, sizing, and portfolio construction

In practice, ADIN treats model-driven screening as core infrastructure and uses human judgment where it has the most leverage: deciding which companies receive capital and how they fit into the portfolio.

The study’s results align with our design. The research supports ADIN’s central thesis: a hybrid of LLM infrastructure and investor judgment is an effective way to identify high-potential startups under modern dealflow volume.

ADIN runs on three connected components:

• a network that supplies differentiated dealflow

• an LLM-native layer that structures and screens companies

• investors who make allocation decisions

Network-led sourcing

Dealflow comes in through scouts. They bring startups from founder, operator, and researcher networks across multiple ecosystems, with a focus on infrastructure, agentic tooling, and new financial rails. That creates a large, thematically concentrated pipeline where an LLM screening layer can work at fund scale.

LLM-driven structuring and screening

Once a company is submitted, ADIN’s system layer applies the same class of methods tested in the study. It:

• standardizes public data and internal notes into a single profile

• embeds and groups startups under clearly defined investment theses

• maintains an updated map of how ventures sit within and across those clusters

Screens and shortlists are generated from this structure, which gives clearer views of each theme and faster identification of companies that warrant investor time.

Investor-led decisions

Capital allocation remains with investors. They review structured views of each thesis area, cluster maps, and the specific companies surfaced by scouts and by the system. They add founder and market context, reference work, and portfolio considerations, then decide what to fund and how to size positions.

The study finds that the best survival and funding outcomes sit where the LLM agent and human analysts agree. ADIN’s pipeline is designed to expose that overlap quickly and make it the center of the investment workflow.

ADIN is a venture platform that pairs an LLM-native infrastructure layer with a global scout network and professional investors. The system automates first-pass analysis, structures company data, and produces investment materials in minutes. Scouts who surface funded startups participate in carry, and investors use the output to make allocation and portfolio decisions.

Learn more at adin.online. If you are interested in participating as a scout or investor, you can apply here

A new peer-reviewed study ran 61,814 startups from Freigeist Capital’s pipeline through an agentic LLM and compared the results to Freigeist’s own analysts.

A human analyst needed around 2 hours for a thesis-driven screen. The LLM agent did the same job in about 13 seconds, a 537× speed-up for the same class of screening.

On screening quality, the agent stayed close to human performance on a standard clustering metric:

• Human analysts: 0.37 Silhouette

• LLM agent: 0.35 Silhouette

On the Calinski–Harabasz index, the LLM scored about 70% higher, which signals tighter, better-separated clusters across more than 61k ventures.

In follow-up data, startups surfaced by the LLM were more likely to survive and more likely to raise follow-on funding than the baseline group, including some that human analysts did not highlight. Ventures selected by both the agent and the analysts showed the strongest survival and funding rates.

Taken together, the study shows that an LLM-driven screening layer can run on real VC dealflow at scale, stay within the performance band of experienced analysts, and surface companies with stronger basic outcomes.

This is the layer ADIN has treated as core from the beginning.

The Freigeist study shows that an LLM-based screening layer can:

• run thesis-driven screens about 537× faster than a human analyst

• stay in the same quality range on Silhouette score

• deliver significantly higher cluster separation on Calinski–Harabasz

• surface startups with higher survival and follow-on funding rates, with the strongest signal where model and human selections overlap

ADIN is built for that hybrid.

• Scouts broaden and diversify the pipeline

• LLM-native infrastructure structures and screens that pipeline at scale

• Investors concentrate on selection, sizing, and portfolio construction

In practice, ADIN treats model-driven screening as core infrastructure and uses human judgment where it has the most leverage: deciding which companies receive capital and how they fit into the portfolio.

The study’s results align with our design. The research supports ADIN’s central thesis: a hybrid of LLM infrastructure and investor judgment is an effective way to identify high-potential startups under modern dealflow volume.

ADIN runs on three connected components:

• a network that supplies differentiated dealflow

• an LLM-native layer that structures and screens companies

• investors who make allocation decisions

Network-led sourcing

Dealflow comes in through scouts. They bring startups from founder, operator, and researcher networks across multiple ecosystems, with a focus on infrastructure, agentic tooling, and new financial rails. That creates a large, thematically concentrated pipeline where an LLM screening layer can work at fund scale.

LLM-driven structuring and screening

Once a company is submitted, ADIN’s system layer applies the same class of methods tested in the study. It:

• standardizes public data and internal notes into a single profile

• embeds and groups startups under clearly defined investment theses

• maintains an updated map of how ventures sit within and across those clusters

Screens and shortlists are generated from this structure, which gives clearer views of each theme and faster identification of companies that warrant investor time.

Investor-led decisions

Capital allocation remains with investors. They review structured views of each thesis area, cluster maps, and the specific companies surfaced by scouts and by the system. They add founder and market context, reference work, and portfolio considerations, then decide what to fund and how to size positions.

The study finds that the best survival and funding outcomes sit where the LLM agent and human analysts agree. ADIN’s pipeline is designed to expose that overlap quickly and make it the center of the investment workflow.

ADIN is a venture platform that pairs an LLM-native infrastructure layer with a global scout network and professional investors. The system automates first-pass analysis, structures company data, and produces investment materials in minutes. Scouts who surface funded startups participate in carry, and investors use the output to make allocation and portfolio decisions.

Learn more at adin.online. If you are interested in participating as a scout or investor, you can apply here