Where AI over data goes wrong
Valid query.
Wrong answer.
Wrong answer.
"Closed deal" means something different in sales than in finance. AI has no way to know. It queries correctly and returns an answer your CFO and your VP Sales would each dispute. They'd both be right.
What we built
A meaning layer that encodes how your organization defines its own metrics. The AI answers questions the way your business thinks, not the way a generic model guesses.
Technically correct.
Organizationally wrong.
Organizationally wrong.
The BI team spent a week on that report. The exec looks at it and says "that's not how we define retention." Everyone is right. The model just didn't know. Executives stop trusting the system. The data lake gets bigger. Confusion breeds disbelief.
What we built
We extract tribal knowledge: how your teams actually define your metrics. We bind it directly to the data. Agents stop guessing. They start answering.
The knowledge lives
in people's heads.
in people's heads.
"Active user." "Healthy account." "Good quarter." The definitions exist, but they're in Slack threads, in the heads of the analyst who built the dashboard three years ago, in undocumented logic nobody wrote down. AI has no access to any of it.
What we built
We sit inside your existing stack and extract that tribal knowledge: structured, bound to the data, and growing smarter with every interaction.
What we do
We build the semantic layer your enterprise data is missing: the encoded understanding of what your data means, how your organization defines its metrics, where the knowledge lives that no schema captures. Once that layer exists, every AI tool you use answers questions the way your business actually thinks. The layer deepens over time. Every correction, every answered question enriches it. It becomes the most accurate representation of how your organization thinks about its data — and it belongs to you.
Intelligence
AI that answers like your organization thinks
When the semantic layer is in place, AI stops returning answers that are technically correct but organizationally wrong. The BI bottleneck breaks. Executives get answers they trust. The data team stops fielding the same questions every week.
Interlochen · 807K contacts scored, 17K priority leads surfaced in 6 days
Ink · relationship intelligence across creative industry contacts and deal flow
Ink · relationship intelligence across creative industry contacts and deal flow
Infrastructure
Pipelines built on meaning, not just structure
The semantic layer informs every pipeline and transformation that runs beneath it. We connect disparate data sources and build the infrastructure on top of the same meaning layer: pipelines that know what the data means, not just what shape it's in.
Major Financial Institution · automated pipeline generation, 80%+ runnable output, 50%+ engineering time saved
Visa / DTCC · multi-source data connectivity & agent fleet deployment · in discovery
Visa / DTCC · multi-source data connectivity & agent fleet deployment · in discovery
How we work
We extract what lives
in people's heads.
in people's heads.
+
The definitions that make AI trustworthy don't live in your schema. They live in the heads of your analysts, the person who built the dashboard three years ago and left, the undocumented logic nobody has time to write down. We extract that knowledge through a structured process and encode it into the semantic layer. What was implicit becomes explicit, machine-readable, and owned by your organization.
Meaning before agents.
Always.
Always.
+
Most AI deployments fail at the same place: they skip the semantic layer because it's the hard part, deploy tools on top of raw data, and get answers that are plausible but wrong. The model isn't broken. It just doesn't know what "churn" means at your company, in your data model, as defined by your VP of Product versus your CFO. The semantic layer solves this. Agents, pipelines, intelligence tools: all come after. This order is not optional.
The longer it runs,
the smarter it gets.
the smarter it gets.
+
The semantic layer is not a one-time build. Every correction refines it. Every question answered deepens it. Stratum accumulates organizational intelligence continuously: new definitions, refined metrics, resolved ambiguities. After two years it is the most precise representation of how your organization understands its data that has ever existed. That asset belongs entirely to you.
We work inside your stack.
Nothing moves, nothing breaks.
Nothing moves, nothing breaks.
+
The semantic layer lives on top of your existing stack. Your Snowflake stays. Your Salesforce stays. Your warehouse, your CRM, your pipelines: nothing moves, nothing gets replaced. We add the meaning layer your data was always missing. No migration. No rip-and-replace. No six-month implementation before the first answer. What changes is whether AI can understand it.
8
Years building data infrastructure at scale
7T+
Rows of data ingested and maintained
700M+
Entities scored and disambiguated
250+
Skills in the semantic layer
Chapter 02 · The Platform
The semantic layer.
And everything it enables.
Stratum is the semantic layer at the center of everything edisyl builds: the encoded understanding of what your data means, how your organization defines its metrics, and where the knowledge lives that no schema captures. Forge builds agents equipped to act on that layer. Lattice orchestrates them as a fleet. The layer is the product. Everything else is what it enables.
“Without a semantic layer, every AI tool you deploy is guessing at what your data means. The right answer is knowable. It just has to be encoded first.”
The architecture
Your data: any source, any structure CRM Warehouse Email Docs APIs On-chain
Forge
Agent framework & specialized tools
Where agents are built and equipped. Each is purpose-trained with domain knowledge and specialized tools. Outputs are quasi-deterministic because agents use tools. They do not see raw data directly.
Specialized toolsSQL, APIs, models
250+ skillsDomain-trained
Memory managementLong-running tasks
Forge
Where agents are built and equipped. Shaped under pressure, with intention.
Lattice
Fleet orchestration & coordination
Coordinates agents as a fleet. Routes tasks, manages context handoffs, tracks the dependency tree. Agents run concurrently. Every action is SQL-traceable.
Fleet plannerTask routing
Dependency treeOrdered execution
Audit trailSQL-traceable
Lattice
An interconnected structure where every node relates to every other.
Stratum
Semantic intelligence & knowledge layer
Encodes how your organization understands its own data: what terms mean, how tables relate, where knowledge is complete and where it is not. The LLM touches interpretation in one controlled phase.
Meaning layerWhat your data means
Coverage mapGaps visible & filled
Continuous ingestionAlways learning
Stratum
Layers of accumulated knowledge. The more the system runs, the deeper and more precise it becomes.
Outputs into your systems CRM write-back Slack briefings Pipeline code Intelligence reports
How a deployed fleet works · illustrative quantifiable outcomes
Scheduled or on-demand
Lattice orchestrated
Intelligence Manager
Owns coverage universe · routes tasks · compiles findings · triggers brief writer
Always on
Analyst 01
Data & scoring
Scores signals 1–5. Filters noise upstream.
Parallel
Analyst 02
Domain specialist
Trained on your context and decision logic.
Persistent memory
Analyst 03
Competitive tracker
Monitors comparable signals. Flags shifts.
Parallel
Analyst 04
Pipeline agent
Generates & validates pipelines. 80%+ runnable.
On trigger
Brief Writer
Formats for decision-makers · cites sources · delivers to inboxes, CRM, Slack
Weekly
↗ Revenue
Priority leads written to CRM
Scored, ranked, routed to the right rep. Daily.
Interlochen: 17K leads in 6 days
↗ Conversion
Automated outreach & follow-up
Right message, right contact, right moment.
Briefings posted to Slack automatically
↗ Efficiency
Pipelines & workflows in production
Generated, validated, deployed. Engineers freed.
Financial institution: 50%+ eng. time saved
Chapter 04 · Proof
Live deployments.
Measurable outcomes.
The same three-layer system configured for what you actually need. Two orientations, both in production. Neither required starting from scratch.
“The architecture is consistent. The problem it solves is yours.”
Growth Track · Interlochen Center for the Arts
807,000 contacts scored.
17,000 priority leads surfaced.
17,000 priority leads surfaced.
Interlochen had over 1.2 million unstructured note fields in HubSpot that no tool had ever touched. We ingested all of it, built a lead scoring model trained on their specific signals, and delivered scored leads directly into HubSpot within six days. Daily write-back continues. Weekly hot-leads briefing posts automatically to Slack.
6 days
Time to first output
AUC 0.734
Model accuracy
807K
Contacts ingested & scored
Daily
Live score cadence
Efficiency Track · Major Financial Institution
DBT pipelines generated.
Engineering hours returned.
Engineering hours returned.
A regulated financial institution needed to automate the generation of DBT transformation pipelines from existing SQL logic — a task their engineering team handled manually at significant cost. Our Streamline tool automated the generation, validation, and iteration loop. 80%+ of output was production-runnable. Engineering time on this class of work dropped by more than half.
80%+
Runnable output rate
50%+
Engineering time saved
Private
Deployment model
Chapter 05 · Foundation
Eight years building
where it was hard.
Most AI companies were built in the last eighteen months. Edisyl was built over eight years on blockchain — the most unstructured, high-velocity, pseudonymous data environment that exists. 700 million entities. Seven trillion rows. Twenty chains. The tools and methodology we built there are what we deploy against your enterprise data today.
IP development by era
2017–2021
Data Foundation
Built ingestion across 20+ blockchains. 700M+ entities scored. Foundational patent granted.
2021–2024
Data IP
Streamline, LiveQuery. Client onboarding compressed from 4 weeks to 2 hours.
Early 2025
First AI Product
Found the accuracy cliff with LLMs. Semantic thesis formed. Built what did not exist.
Late 2025
Integrated AI System
Forge, Lattice, Stratum. Enterprise deployments live. First measurable outcomes delivered.
2026+
Enterprise Scale
Private deployment. Any data source. Continuous learning inside the client environment.
Patent
Patent US10706472B1 · Active · Filed 2019 · Granted 2020 · Expires 2039
Systems and Methods for Analysis of Digital Asset Development and Transaction Behaviors
Status Active
Expires 2039
Proprietary tools, applicable anywhere
Streamline
SQL-to-pipeline compiler
Transforms query logic into production-grade transformation pipelines. Drives automated DBT generation: 80%+ runnable output, 50%+ engineering time reduction.
LiveQuery
SQL-to-any-API translation
Agents query live data using familiar SQL syntax, no custom connectors required per source. The bridge between structured query logic and live production systems.
Scoring Methodology
Patent-backed entity scoring
700M+ entities scored and disambiguated across 20+ chains. The same methodology applies to any enterprise entity graph: customers, contacts, accounts.
edisyl · 2026
Talk is cheap.
Tell us what you have and what you're trying to accomplish. We'll take it from there.