One evidence layer under every transaction document.
Khumo is one system, not a set of tools. Everything a company's documents contain becomes verified, cited facts, and every deliverable is built from those facts. This page describes the system and why it gets better with use.
The system
A customer creates a company once and adds its documents once. From then on every action runs against the same evidence: reading the data room, checking every figure, screening against criteria, and filling teasers, information memoranda, memos and IC decks in the customer's own design. Inside the application there are no products; the four modules below are how we describe the engineering.
The asset is not document generation, which every model will offer. The asset is the structured, permissioned and steadily improving layer of evidence underneath the documents, and the workflow built on it.
Inventory, classification, duplicates, versions, coverage against what a process needs, and a content-based map of the data room that lets inexpensive models do most of the work.
Extraction with page and cell citations, contradictions between documents, deterministic computation of growth, margins and 31 KPIs, and search across all documents for clauses and figures.
Sale readiness for advisers, investment criteria for funds, risks by category, the request to the client and the questions for management, with a verdict a partner can defend.
The customer's own master template, mapped once and filled from confirmed facts, with native tables and charts, versions and approval.
Why it improves with use
Every serious competitor has the same models. What accumulates at Khumo is different:
- Lineage at the level of the individual figure: page, cell and approver, across the life of the deal.
- A DACH transaction vocabulary: German notation, HGB and IFRS structures, scanned statements, owner-managed perimeters, contract clause types.
- A benchmark corpus with hand-verified answers, run on every prompt and model change and extended with closed deals from founding customers.
- Corrections as signal: what the machine wrote, what the analyst changed, and whether the change repeats.
- Workflow integration: the corporate-design template, the request page and the approval step make Khumo part of how a house works.
Where we stand
Two products are live with users: a buy-side screening application and a document generator that already fills a client's own information-memorandum master. A sell-side pre-screen has been validated on synthetic and real material and is being validated on closed deals with founding customers. The unified platform that puts one evidence layer under all of it is being built in a fixed sequence with a validation gate at every phase.
We publish prices, sign founding customers on written commitments, and do not describe demos as traction.
Who we are
Khumo Automation UG, Munich, a spin-off from TUM and LMU and a member of TUM Venture Labs.