AI that proves
it's right.
Every AI platform gives you its best guess. For the decisions that define careers, move markets, and stand up in court, guessing isn't good enough. Kepler makes AI verifiably correct.
Every answer can be traced to its source.
Like planetary orbits, every Kepler output follows deterministic laws.
"Probably right" is not "provably right."
Large language models are built on probabilities, not certainty. There is always an element of randomness that cannot be eliminated. Ask for a revenue figure twice and you may get two different numbers, neither with a citation.
But the real problem isn't just that AI gets things wrong sometimes. It's that even when AI gets things right, you can't prove it. You can't trace the number. You can't defend it in a boardroom or a courtroom. Trust requires proof.
AI handles the language. Code handles the data. They never cross.
Four layers, one principle: every answer is traceable. AI interprets your intent. Deterministic code retrieves and computes the data. The layers communicate through structured contracts, never raw model output.
Every team works differently. The Personalization Layer stores templates, saved queries, output preferences, and workflow configurations so Kepler adapts to how your team thinks, not the other way around.
The AI layer interprets your question, decomposes it into sub-tasks, and builds an execution plan. It understands language and context. It never retrieves data or computes numbers directly.
> "Compare AAPL and MSFT gross margins, last 4 quarters"
1. resolve_entities(["AAPL", "MSFT"])
2. fetch_metric("gross_margin", periods=4)
3. compute_comparison(method="side_by_side")
4. format_output(type="table")
5. attach_citations()A structured knowledge layer that maps entities, relationships, and data lineage. Every concept is resolved against a shared ontology so "revenue," "top line," and "total sales" all point to the same verified number.
Code retrieves and computes. Every number comes from a verified source with a citation. Every calculation is reproducible. No probabilities, no randomness. Same input, same output, every time.
Kepler Finance
Live and in production. Purpose-built for investment banking and financial analysis. AI handles the complexity of your question. Deterministic code handles the data. SEC filings, earnings transcripts, 10-Ks, 10-Qs, and market data are indexed and ready to query from day one. All traceable.
All the data, already there
SEC filings, earnings transcripts, real-time market feeds, and regulatory documents. Normalized, structured, and indexed with full provenance so you never start from scratch.
Analysis with full traceability
Company fundamentals, peer comparisons, debt structures, revenue breakdowns. Every number sourced to the original filing, the page, the line item. Click any figure to see where it came from.
Deliverables you can defend
Output directly to Excel and PowerPoint with every data point verified before it leaves Kepler. What used to take an analyst hours of cross-checking now takes seconds.
Kepler Platform Coming Soon
Everything we built for finance, we built to be industry-agnostic. Kepler Platform opens the full platform to any domain. Connect your own data, define your own workflows, and deploy AI agents that use each system for what it's best at. Your industry. Our architecture. Every answer traceable and defensible by default.
Connect any data source
Databases, APIs, document stores, internal knowledge bases. Kepler indexes and normalizes them with full provenance, regardless of industry or format.
Define your workflows
Build the logic your domain requires. Compliance checks, risk analysis, research synthesis, operational reporting. You define the rules. Kepler enforces traceability.
Deploy verifiable AI agents
Every agent you build inherits the same separation: code handles data, AI handles language. Your users get answers they can trace, verify, and defend.
AI is at its best when you can see how it works. Kepler Platform gives you the same deterministic architecture we use internally. The same separation of code and language. The same audit trail. The same guarantee: if Kepler says it, you can prove it. That is not a feature. That is the foundation.
Built by engineers who've spent
their careers getting answers right.
The Kepler team comes from Palantir, Citadel, Meta, Stanford, and EPFL. Together, they've built and operated AI in some of the most critical environments in the world, from defense and intelligence to quantitative finance. Kepler is the product of that experience: AI you can trust and verify.
Vinoo Ganesh
7 years at Palantir leading storage and compute platforms. Built Veraset to $15M ARR (acquired). Citadel's first Head of Business Engineering.
Dr. John McRaven
11 years at Palantir. Built the analytics engine behind $100M+ contracts. Created Palantir's first AI platform. PhD in Physics.
Sara Kromwijk
11 years at Palantir. Architected product enabling Palantir's transition from per-customer to vertical-based FDE engagements. MSc in CS from EPFL.
Susannah Meyer
Senior engineer at Meta on big data systems. Head of Engineering at Enigma Labs, scaling to 300k+ users. BS in CS from Stanford.
Rui Zhang
10+ years as a business manager running $100M+ businesses. PE and portfolio company operations at a $12B private equity firm.
Your rules, encoded permanently.
Kepler learns at every level: analyst, desk, vertical, firm. It remembers corrections forever. No re-teaching next quarter.
Kepler persists preferences at the individual, team, desk, and firm level. Correct a treatment once (EV inputs, cash adjustments, debt classification) and it sticks across every future query.
Skills are deterministic, auditable workflows: EV calculations, trading comps, cap tables. Change how diluted shares are treated, and Kepler uses that rule every time. No black boxes.
Company-specific modeling nuances get encoded and reused. FIG-specific rules like corporate vs. financing debt, loan-book breakdowns, and digital vs. traditional banks are baked into how comps and screens run.
Kepler ingests your comps files and models, then flags outliers and likely errors: weird multiples, broken formulas, stale inputs. Analysts know exactly where to focus.
Every learned rule is visible and editable. Review what Kepler remembers, revise what's wrong, and delete what's outdated. Full transparency into how your preferences shape outputs.
Specify what matters: prior-model numbers, geo-risk exposures, sector-specific filings. Kepler structures just that. Not every document. Just the ones your desk actually uses.
Kepler auto-builds buyer profiles, strategy summaries, and company overviews from presentations and reports. Each line is cited back to the original source so your team can trust it in a pitch.
Structured tables, narrative summaries, raw data, or Excel workbooks with formulas intact. Configure how every answer is returned: by analyst, by workflow, or by deliverable type.
Upload a deck or book and Kepler audits your target lists, flags gaps in your comps, and suggests missing buyers or investors. A second opinion before you go to market.
Quick-check mode for speed on routine screens, or full audit trail with source links for compliance-critical queries. Set the default per desk, override per task.
From due diligence to board reports. Answers you can defend.
Investment Banking
Trading comps, precedent transactions, buyer profiles, and competitive landscapes built from source filings. Every multiple traces to a specific line item. Change how you treat convertible debt or pension liabilities, and the entire comp set recalculates. Pitch books assemble with every data point sourced so your MD can defend the page in front of a board.
Equity Research
Revenue segment breakdowns, margin assumptions, and share counts pulled directly from 10-Qs and earnings transcripts. When a company restates or reports, your model updates with the new numbers and flags what changed.
Hedge Funds
Screen thousands of companies on financial metrics, ownership changes, or sector-specific signals with data verified at the source. Fewer false positives. No missed opportunities because a data vendor lagged a filing by two weeks.
Private Equity
Cross-reference years of filings, footnotes, and disclosures to surface inconsistencies, restatements, and off-balance-sheet exposures. The kind of pattern recognition that takes a team weeks, done in minutes with full citations.
Compliance
Every output has complete data lineage from conclusion back to source document, page, and paragraph. When a regulator or auditor asks where a number came from, the answer is already there.
Accounting & Tax
Cross-reference tax disclosures, effective tax rate reconciliations, deferred tax assets, and ASC 740 footnotes across years of filings. Surface transfer pricing exposures, NOL carryforwards, and valuation allowance changes with citations to the exact disclosure.
Enterprise-grade security at every layer.
Every output can be traced, explained, and proven.
Security is a baseline architectural requirement
Security is embedded into every layer of the platform, from infrastructure to application logic.
Strict data isolation and boundary enforcement
Your data is stored in siloed environments, fully isolated from other customers at every level of the stack.
Least-privilege access and controlled permissions
Every user, process, and service operates with only the access required. Permissions are scoped, logged, and reviewable.
Full auditability of system behavior
Every query, calculation, and data access is logged with complete lineage. Trace any output back to its source.
Deterministic and reproducible outputs
Same input, same output, every time. Results can be independently verified and reproduced on demand.
Designed to withstand scrutiny
Built for regulated environments where auditors, compliance teams, and regulators require full transparency.
Backed by the engineers & operators behind the most consequential technology of our generation.
Help us build AI the world can trust.
Deep technical problems. Industry-agnostic infrastructure. Everything you build matters.
See open rolesAI that proves
it's right.
Request a demo to experience the difference between AI that is probably right and AI that is provably right.