Lodestone

Test your hypothesis on your private haystack

lodestone.it Pitch deck

01 · Problem / The status quo

Billion-dollar decisions still rely on manual document review. The needle they miss breaks the deal

75% → 20%

Attorneys believed they had retrieved 75%+ of relevant documents; in reality they found about 20%.

Roitblat et al., Document Categorization in Legal Electronic Discovery

35 → 88

Average length of an M&A contract, in single-spaced pages, has more than doubled over twenty years.

Coates, J. C., Why Have M&A Contracts Grown? Evidence from Twenty Years of Deals

The problem

In due diligence, finding a deal-breaker "needle in the haystack" across thousands of pages is very hard for humans. Existing AI tools fail here:

  1. They lack the accuracy to find these details without hallucinating.
  2. They rely on external API, making secure handling of NDA-bound data impossible.

NOT(the solution)

Hire more burned-out interns? Keyword search… seriously? The alternative? AI chatbots hallucinate, their context window is too small, and there is no provable way to trust the output. The last option is wrestling with legacy software with a brutal learning curve and features that don't justify the pain

02 · Solution / What Lodestone is

The adversarial due diligence engine you own

  1. Lodestone minimizes hallucinations by simulating a court case with AI agents. It lets you test high-stakes hypotheses against massive amounts of private company data in minutes, with full traceability.
  2. It can run locally in an air-gapped environment for NDA-bound work, or simply as a SaaS when flexibility matters.

Adversarial research

A Red agent and a Blue agent respectively attack and defend your hypothesis. A judge provides a verdict with precise document citations supporting each argument.

Offline or in the Cloud

If the client requires complete data secrecy, it can run everything offline, off the grid, on its own machine.

Thousands of documents

Lodestone's architecture is built for scale. All the core features experience no slowdown as the document corpus grows.

02.01 · Solution / Court trial

Test your hypotheses

Ingest documents, make an hypothesis, let adversarial agents test it against your corpus.

Ingest

Drop in contracts, filings, or reports. Lodestone parses, chunks, and embeds every document.

Hypothesis

State a claim about your documents and let Lodestone gather the documents that support or refute it.

Debate

Using the relevant documents, the Red Agent attacks the hypothesis while the Blue one defends it across multiple rounds.

Supported
Mixed
Unsupported

Verdict

A Judge weighs both sides and delivers a structured ruling: confidence score, cited evidence, and dissenting arguments.

02.02 · Solution / Search & chat

Ask a question, audit the sources

Builds upon the standard RAG and chat features that you can find elsewhere

Ask

Ask a question over the whole corpus or just a selection of files.

Agent searches

Agent searches through the documents, finding the most relevant and delivering a response with precise citations.

Audit

Check the relevant sources, by clicking on the agent's citations to open the original passages.

02.03 · Solution / Data ownership

Built for NDA-bound diligence and classified environments

Choose where Lodestone runs. Every deployment offers the same features.

Air-gapped Hybrid Cloud

Air-gapped

Complete network isolation. Everything runs offline with local models.

  • Local models only
  • Classified environments
  • Document secrecy
Flexible

Hybrid

The middle ground. Clients can run the core pipeline on their infrastructure while offloading specific tasks to external APIs.

  • Granular control
  • Mix local & remote models

Cloud

Traditional managed deployment on our infrastructure. We handle uptime, scaling, and updates.

  • Managed infrastructure
  • Auto scaling
  • Low maintenance

03 · Market / Targets, pricing, TAM

A multi-billion document-heavy market

Customer Profile

  • Due diligence teams
  • Law firms
  • M&A advisors

Document-heavy practices where a single missed clause can break a deal and where data can't leave the building.

Pricing

On-prem license

  • SMB $30k upfront · $5k/yr maintenance
  • Enterprise $300k upfront · $50k/yr maintenance

SaaS pay-per-use

  • Teams Expected $100 / month / team (avg.)

Total addressable market

$3.5B upfront
$1.5B ARR

Enterprise on-prem

1k deployments

$300M upfront + $50M ARR

SMB on-prem

100k deployments

$3B upfront + $500M ARR

Teams SaaS

1M team accounts

$1B ARR

04 · Traction / Tested & ready for use

Fully functional product, deployable today

Installable on a company's local server, or usable online with unclassified material.

Testers

3 companies testing

  • 1 law firm
  • 1 compliance company
  • 1 engineering company

Future clients

10 law firms in talks

Connected with law firms evaluating Lodestone for due diligence and contract review.

05 · Team / The founders

Together on AI research, startups, and life

PC

Pietro Casavecchia · CEO

Theoretical CS background, logic & NLP. I love breaking down everything to atoms and abstracting it up. Not only a theory guy, I want to be pragmatic, really like working with people and enjoying all aspects of life (e.g. BJJ)

  • Ex-Tech Lead @Oxford AI-inference startup
  • Researcher @ACLAI Lab — 5 projects, 6+ papers (3 as speaker)
  • Game dev & Creator @Enakeli (by Aryarna)
  • Fellow @Silicon Valley Fellowship · Member @Confindustria · Tech advisor @RANRAN
PB

Pietro Bellodi · CTO bellux.dev

I enjoy experimenting with local LLMs and coming up with custom automations. Currently part of ACLAI Lab and contributing to open source in the free time.

  • HFarm AI hackaton winner
  • Research grant for D&DGroup to build an offline RAG
  • Open source contributor (LocalAIME, Perplex, Java2Smali)
  • At 14, disclosed a critical API flaw to Sketchware's founders and joined the team for 2 years as moderator and security contributor

Known each other

5 years

Met at University of Ferrara, both full-time on Lodestone since December 2025.

Adversarial due diligence engine you own

lodestone.it