How our AI works
AI agents collect and read your fund documents. Hundreds of financial rules check every figure, experts review anything the rules flag, and every number stays linked to the page it came from.
At a glance
Built for: capital allocators in private markets
Accuracy: 98% out of the box, 99.99% after expert review
Checks: hundreds of private markets financial rules, run on every figure
Review: specialists review every figure a rule flags
Traceability: every number links to its source page, with its full change history
Independent check: IRR, TVPI and DPI recalculated from your own cash flows
Your data: never mixed with other clients' data, never used to train AI models
AI models: not tied to one provider
Why a general AI tool isn't enough
General-purpose AI is remarkable, and Tetrix uses foundation models every day. But a tool built for everyone can't go as deep as private markets need. Capital allocators need a system built for their industry, with AI inside it:
A system of record that keeps your portfolio data consistent across quarters and years
Workflows that cover the whole job, from document to decision
Connections to GP portals, inboxes and data rooms, many of which have no API
Each client's data kept separate, with access set by role
An audit trail on every value
“Focus compounds. We spend every hour on one kind of customer’s problems, so every quarter we get better at them. A tool built for everyone can’t do that.”
Naunidh Singh Bhalla, Co-Founder & CTO
Why vertical AI wins in private markets
From document to verified data
Collection
AI agents log into your GP portals and data rooms, including those behind two-factor login, and collect each new document the moment it posts. They also sweep every portal periodically, pick up documents from your inbox, and keep working when a portal changes its layout, which is where scripted scrapers break.
Extraction
A reasoning model reads each document the way an analyst would, including tables, footnotes and commentary. It works on PDF, Excel, Word and screenshots, with no templates to set up, and returns data at investor, fund and company level.
Structuring
Each document is classified by what's in it, not by the portal's label, and every figure is put into the same format as the rest of your portfolio. Company names are matched across GPs, so one company reported under three names becomes one record.
Validation
Language models can be confidently wrong, so nothing is trusted until it passes hundreds of financial rules. For example, a capital account's closing balance has to equal its opening balance plus the quarter's activity. Anything that fails a rule goes to a finance specialist, who checks it against the source document and corrects it.
Release
Nothing reaches your dashboards until it has been checked. Each number keeps its source page and its full history: who changed it, when, and which rule flagged it. You can override any value yourself, and that's recorded too.
How accurate is Tetrix?
98% out of the box, after the rules check. 99.99% after expert review. Capital calls and distributions come through in about an hour.
In private markets, 99% correct is still wrong. A NAV or an exposure figure has to be right, and small errors compound across thousands of documents. Generating an answer is cheap. Checking it is what takes time, so Tetrix makes checking any number take one click.
Every return, calculated twice
Tetrix records IRR, TVPI and DPI as your GP reported them, then recalculates each one from your own cash flows. When the two differ you see a flag, and you can open the calculation to see every input and every derivation.
Answers you can check
When you ask Tetrix a question, the answer comes from your stored, verified portfolio data, not from the model's general knowledge. Every answer cites its sources, and one click opens the page. Because the data is already extracted and structured, answers come back straight away instead of re-reading documents each time.
Prefer to work in your own AI tools? Connect them to Tetrix through our MCP server.
Built for institutions
Answers you can defend. “The model said so” isn’t an answer an auditor, investment committee or board will accept. Every value in Tetrix traces back to its source, its reviewer and any change made to it.
Your portfolio, only yours. Each client sees only its own portfolio, and one client's data is never mixed with another's.
A team that moves fast. When a client added a new sub-strategy to track across three reporting periods, we turned it around in 24 hours.
Your data, our models and pipelines
Not tied to one model. We use the model that performs best for each task, and switch as models improve.
Never used for training. Your data is never used to train AI models.
Identifying details removed. Wherever possible, we strip identifying details from anything sent to a model.
Yours to take. You can export your data at any time, and it's deleted if you leave.
Visit the trust center
Three questions to ask any AI vendor, including us
Interrogate the verification layer, not just the model.
How is accuracy measured, on what data, and how often is it re-tested? Who reviews the results when the model gets it wrong?
Validate workflows, not tasks.
Most tools look good on one document. Test the whole job: documents arriving through every channel, checked across funds and periods, and ending in a number your investment committee can trust.
Test the speed of the vendor.
Ask what they shipped most recently and how long a typical request takes, then check both with a reference customer.
Read Naunidh's full series on AI:
Common questions
Can Tetrix's AI make up a number?
Tetrix's AI, like any language model, can produce a wrong value, which is why Tetrix doesn't rely on the model alone. Every figure is checked against hundreds of financial rules, flagged figures go to a Tetrix expert, and every number links to its source page so you can verify it yourself.
Why does accuracy matter so much in private markets?
In private markets, 99% correct is still wrong, because a wrong NAV or exposure figure can become a compliance problem and small errors compound across thousands of documents. That's why Tetrix checks every figure against hundreds of financial rules, has Tetrix experts review what gets flagged and links every number to its source page so you can verify it in one click.
Does Tetrix use our data to train AI models?
No. Tetrix never uses client data to train AI models, and never mixes one client's data with another's. Anything sent to an AI model has identifying details removed first. You can export your data at any time, and it's deleted if you leave.
Which AI models does Tetrix use?
Tetrix isn't tied to one AI model provider. It uses the model that performs best for each task and switches as models improve. Your data is never used to train those models, and identifying details are removed from anything sent to one wherever possible.
Can't we just use a general AI tool?
General AI tools such as ChatGPT and Claude are built to be useful across every industry, so they aren't built for the depth private markets operations need. Tetrix uses foundation models itself, but it also keeps your portfolio data consistent across quarters, logs into GP portals that have no API, checks every figure against hundreds of financial rules, has Tetrix experts review what gets flagged and keeps an audit trail on every value. You can still use those tools on top of your Tetrix data through our MCP server.
See all FAQs
Bring us your messiest fund document.
In a demo, we'll run it through every stage live, from collection to a checked, cited answer.
Book a Demo