MetisLLM ATHENA AIC

Building the forest
from the trees.

MetisLLM reads a fund’s entire inbound research flow and resolves it into the standing view across every desk — what the street is positioned for, what changed this week, who is offside, and the page behind every claim.

How it holds up
FIG. 01 — THE EVIDENCE GRAPH ONE FUND’S RESEARCH FLOW, RESOLVED SUPPORTS OPPOSES UNDATED
The job

Four questions, every Monday morning.

01

What is the street actually positioned for?

Not one note summarised — the standing view across every desk you read, with the split shown.

02

What changed since last week?

Which views gained or lost a firm, and which desks changed side.

03

Who is offside?

Named — on both sides, with how many calls each firm has behind it.

04

Can I verify it before I act on it?

Every line resolves to a firm, a date and a page number in the source document.

Measured, not promised

Every number traces to a firm, a date and a page — at any scale.

This is not retrieval. A chat assistant answers from a sample of your documents, recall falls as the corpus grows, and the misses are silent. MetisLLM resolves every document into positions on a shared set of views — so each report added lengthens a history, tightens a count, and sharpens a disagreement.

We delete more than we show.

A desk restating last month’s published print is not forecasting. Turning that one guard on cut our own scored forecast population by 85% — and every implausible extraction we prune is listed on screen by firm, month and value, not silently dropped.

Consensus carries a vintage.

Eleven desks agreeing this week is a crowded trade. Eleven desks agreeing across two years is not a trade at all. Every position is dated off publication timestamps, so the whole map can be replayed to any past date.

Scales are the measured capacity of the live system, not a projection.

How it works

The complexity we manage.

The market at any fidelity. The top level is a board you can scan; every level under it is rebuilt from the documents themselves.

One drill-down: the board of grouped views; a callout opens US Duration into its sub-views; a second callout opens 30Y into the evidence field
ONE DRILL — BOARD → VIEW → EVIDENCE. EVERY LAYER IS RESOLVED FROM THE ONE BELOW IT.
Case study

Eighteen months of one fund’s research flow.

We ran the system end to end on a real corpus: 962 reports from 20 institutions — 26,368 pages, 257,548 dated and sourced facts, resolved into 156 standing views across seven asset classes. Then we rebuilt what it would have shown on past dates. By April 2026 it showed nine of eleven desks long US duration while yields had already turned — the street adding to a losing position for six months, with the two dissenting desks’ written reasoning dated and page-cited.

Read the full case study (PDF)

Banks anonymised; outcomes checked against public FRED series. Episodes selected with outcomes known. Not investment advice.

The team

Built by people who sat on both sides of the research.

Jon Centurino

Jon Centurino

Co-founder & CEO

Former Managing Partner at Torrey Pointe Capital and Senior Managing Director at Cerberus Capital. Previously CIO at Ally Financial managing global portfolios.

LinkedIn ↗
Henry Gilbert

Henry Gilbert

Founding Engineer

Co-founder of Athena AIC. Ph.D. in Computer Science from Vanderbilt, specialising in LLMs and financial applications. Trading systems, document intelligence, and production ML infrastructure.

LinkedIn ↗
Michael Sandborn

Michael Sandborn

Founding Engineer

Co-founder of Athena AIC. Ph.D. in Computer Science from Vanderbilt with research funded by DARPA and the NSF. Trading systems, malware analysis, and design optimisation.

LinkedIn ↗
Alexy Skoutnev

Alexy Skoutnev

Founding Engineer

Founding engineer at Athena AIC, building AI-driven investment research platforms. M.S. in Computer Science from Vanderbilt, specialising in LLMs for investments.

LinkedIn ↗
Access

See it on your own research flow.

A pilot starts with your book: you name the positions you hold and the desks you read, and we report coverage honestly on day one.