“
Haystac has completely changed how we manage documents. What used to take hours of manual processing is now done in minutes—with greater accuracy.
Operations Manager
Financial Services
Haystac Platform / Chicago
Document correlation for high-volume inbound content.
Chicago automatically separates, identifies, and organizes documents arriving in mixed batches so downstream systems start with the right document in the right context.
Auto-separate
Break mixed files into distinct business documents with minimal setup.
Auto-identify
Classify content by meaning, not just rigid templates or keywords.
Ready for downstream workflows
Feed parsing, routing, and automation layers with cleaner, structured input.
Where Chicago Fits
Step 1: Organize content
Turn bundled inbound files into separated, labeled, workflow-ready documents.
Built on embeddings
Chicago uses embeddings to understand semantic similarity and improve classification fidelity.
Feeds Nashville and Orion
Once documents are separated and identified, they can move into parsing and insight generation.
What Chicago does
Chicago addresses the digital mailroom problem by separating batches of inbound documents and identifying what each document is before further processing occurs.
Separate mixed files
Split a single uploaded file into discrete business documents such as invoices, claims, statements, forms, and supporting materials.
Classify by meaning
Use semantic understanding to distinguish similar-looking documents and reduce manual sorting effort.
Prepare for downstream processing
Create cleaner inputs for parsing, adjudication, workflow automation, and generative insight.
Correlation-first
Organize content before extraction or decisioning begins.
Embeddings-based
Capture meaning and context beyond keyword matching.
Minimal setup
Reduce manual effort required to create classification-ready workflows.
Platform-native
Feeds directly into Nashville, Orion, and Polaris.
How Chicago works
Chicago is designed for high-volume environments where documents arrive in bundles, packets, or mixed uploads and need to be organized before business logic can be applied.
Step 1
Receive inbound content
Files enter through a host application, workflow layer, or upstream intake process.
Step 2
Separate documents
Chicago detects document boundaries inside mixed content and breaks them apart.
Step 3
Classify each item
Each separated document is identified and labeled using semantic classification.
Step 4
Return structured output
Results are passed back to the host or routed into downstream Haystac services.
Key capabilities
Chicago is built for noisy, inconsistent, real-world inbound content.
- Auto-separate documents contained within a single uploaded file or packet.
- Semantic classification using embeddings-based understanding.
- Supports digital mailroom and intake-heavy workflows.
- Returns structured JSON-ready correlation results to the host environment.
Best-fit use cases
Chicago is especially valuable anywhere inbound documents arrive mixed together.
- Claims packets containing multiple insurance documents.
- Loan files with applications, statements, IDs, and supporting records.
- Government intake scenarios with mixed forms and attachments.
- Transportation and logistics workflows with shippers, invoices, and bills of lading.
Explore the platform
Chicago is the correlation layer within the broader Haystac platform.
Start with cleaner inputs.
Chicago helps teams organize inbound content before extraction, decisioning, and automation begin.
Intelligent document processing for regulated enterprises
The AI your business wants. The control your regulators require.
Haystac is a fully containerized AI platform that deploys behind your firewall — on-prem, in your private cloud, or hybrid. Built for banks, insurers, hospitals, and agencies that need to classify, extract from, reason over, and act on the documents that keep getting flagged by security.
No data egressYour documents never leave
No 18-month buildDeploys in weeks
No CISO escalationNIST 800-53 aligned
haystac.local · Mailroom · Live intake
Inbound queue Live
Mortgage_2842.pdf
Sorting
ClaimForm_FB-9183.pdf
Extracting
PolicyDoc_HC-4421.pdf
Done
PA_Request_22834.pdf
Queued
SAR_Q3_BNK-1729.pdf
Queued
Underwriting_pkt_993.pdf
Queued
OmniSuite pipeline280ms median
Sort
Chicago
Classify mixed batches by meaning
Extract
Nashville
Parse fields, tables, handwriting
Answer
Orion
Grounded RAG with citations
Act
Polaris
Trigger workflow & route
Live traceMortgage_2842.pdf
SortClassified as Loan File12ms
Extract14 fields · borrower, DTI, LTV, income86ms
AnswerMatches underwriting policy 98%159ms
ActRouted to Underwriting · S. Chen23ms
Today’s answers
“Are there indicators of fraud in claim FB-9183?”
Three signals match SIU referral criteria: late submission relative to incident date, prior carrier history with similar claim, and adjuster note describing inconsistency between damage photos and claimant statement.
FormA p.2
Notes 03-12
Photos 4-7
SIU §4.2
12,847
Documents today
94%
Straight-through
6
Escalated to humans
280ms
Median latency
Recognized · Compliant · Proven
Gartner 2026 — DSLM
NIST 800-53
170+ patents
ABBYY · Kofax · Lexmark veterans
The status quo
You’ve been told you have three options. None of them work for regulated industries.
The net effect
90%+
of regulated-industry data sits locked and unusable for AI — not because the AI isn’t good enough, but because the AI is in the wrong place.
There’s a fourth option.
Run modern AI inside your data center.
Haystac is a fully containerized AI platform that deploys behind your firewall. Your documents stay where they are. The models train on your content. Every answer is traceable to a source document you already own. NIST 800-53 aligned out of the box.
Your data
Stays inside your environment.By architecture, not policy.
Your models
Trained on your content.Your terminology, your context.
Your control
Governance, traceability, compliance.Built in, not bolted on.
OmniSuite™ pipeline
From inbound document to executed action.
Four products. One pipeline. Sort · Extract · Answer · Act — each independently deployable, all designed to work together.
Sort
Chicago
Classify & route
Inbound documents grouped by meaning, not templates. Mailroom triage, claims intake, case files — classified before they hit a human queue.
Queue
Classifier
Sort
98% confidence
Bins
Loan files7
Claims12
Policies3
SAR1
Extract
Nashville
Multi-modal parsing
Pull every field from forms, tables, handwriting, irregular layouts — without OCR brittleness. Outputs are JSON, ready for your systems.
A. Patel
$486,000
0.32
A.Patel
borrowerA. Patel
loan_amount$486,000
DTI0.32
income$148,400
signaturehandwritten ✓
Answer
Orion
Generative insight, grounded
Get answers from your own documents. Every response cited. Every reasoning chain auditable. Bias toward retrievable content reduces hallucination.
Q
Does loan 2842 match underwriting policy?
DTI 32% sits within the 36% policy ceiling1. Income verified against W-2 and bank statements2. No exceptions flagged3.
Sources
1LoanApp_2842.pdfp.4
2W2_2024.pdftable 2
3Underwriting_Policy.md§3.1
Act
Polaris
Workflow automation
Approve, route, escalate, generate — multi-step workflows with a full audit trail. Turns content-driven understanding into operational follow-through.
Loan File · classified
DTI ≤ 36%?
Yes
Run policy check
Auto-approve
No
Escalate to L2
Notify reviewer
Audit log · 4 entriesall on-prem ✓
Benefits
Six things that change the day you deploy.
Most AI rollouts in regulated industries die in security review. Haystac changes the inputs to that conversation — and the conversation that follows.
01
Use AI on the documents that have been off-limits.
Claims, charts, loan files, case files — finally usable.
02
Pass the audit on day one.
Every answer cited to source. NIST 800-53 out of the box.
03
Deploy in weeks, not 18 months.
Containerized, pre-hardened, ready to install behind your firewall.
04
Cut manual document handling.
Sort, extract, answer, route — without manual handoffs for in-policy work.
05
Keep your existing stack.
REST API. Runs alongside your ECM, CRM, ERP, and case systems.
06
Scale without rebuilding.
Same architecture from 500K to 50M+ pages a year.
Wall of love
From the people who run regulated operations.
“
We reduced document processing time by 80% after implementing Haystac. The AI-powered classification and extraction features have been game-changers for our team.
Director of Compliance
Healthcare
“
Our workflow was drowning in paperwork before Haystac. Now we’ve automated data extraction and improved efficiency across departments.
CIO
Logistics & Supply Chain
“
With Haystac we eliminated manual errors in invoice processing, saving thousands in operational costs each quarter.
Accounts Payable Manager
Retail
Industries
Built for the industries that can’t afford to compromise.
Banking
Banking
Process loan files, KYC packets, and trade docs without leaving your data center. Decisions cited to source.
Loan files
KYC packets
Trade confirmations
SAR reports
Loan_2842 · A. Patel
Auto-approved
DTI
32%
LTV
78%
FICO
742
All policy thresholds met. Income verified. Cited to:
LoanApp p.4
W2_2024.pdf
Policy §3.1
BankStmt Q4
4 documents · 14 fields extracted
✓ Audit log written
Insurance
Insurance
Triage claims, automate underwriting reviews, and read every policy without exposing PII.
Claim files
Policy docs
Adjuster notes
Coverage rules
Claim FB-9183 · auto bodily injury
SIU Referral
Fraud signals detected · 3
!
Late submission · 14 days post-incident
FormA p.2
!
Prior carrier history · similar claim 2023
Notes 03-12
!
Photo / statement mismatch on impact angle
Photos 4-7
Recommendation
Refer to SIU
87% confidence
Healthcare
Healthcare
Extract from charts and prior authorizations under HIPAA, on infrastructure you control.
Patient charts
Prior auth
Lab results
Discharge summaries
PA-7841 · cardiac MRI
HIPAA
Patient
nameredacted
dobredacted
mrn****-2841
Clinical
troponin0.42 ng/mL
EF38%
prior MIyes (2023)
Meets criteria
Cited to Plan §H-12, lab table
Government
Government
FedRAMP-aligned AI for case files, FOIA responses, and benefits adjudication.
Case files
FOIA requests
Benefits forms
Permits
2024-FOIA-0842 · agency contracts
FOUO
Page 3 of 47 · auto-redacted
Subject
Contract award FY24-Q3
Awardee
[Redacted under Exemption (b)(4)]
SSN
PoC
Address
Amount
$1,284,000 (releasable)
14 redactions applied · (b)(4), (b)(6), (b)(7)(C)
✓ ready for response
In production
Four real questions Haystac answers.
Insurance · Fraud detection
“Are there indicators of fraud in this claim file?”
Haystac reviews claim forms, adjuster notes, photos, and prior history. Returns a structured summary with cited evidence — ready for an SIU referral or routine close.
Banking · Regulatory change
“Which of our procedures need to change under this new regulation?”
Haystac analyzes policies, identifies impacted processes, produces a traceable impact assessment — with line-level citations to both the regulation and the affected procedure.
Healthcare · Coverage decisions
“Does this treatment qualify under the patient’s plan?”
Haystac evaluates clinical documentation, coverage rules, and payer guidelines. Returns a defensible coverage decision with citations to the exact policy clauses.
Operations · Straight-through
“Can we auto-process this in-policy claim?”
Haystac classifies, extracts, validates against business rules, and routes. Escalates exceptions to a human queue. Logs every step for the audit trail.
Why trust us
Built by the people who built this market the first time.
The Haystac team led or founded the companies that defined every IDP wave before this one — ABBYY, Kofax, Sapiens, Solaris, ILOG. Now they’re building the platform they always wished existed: domain-specific AI that runs inside regulated environments, not around them.
1990s
ABBYY · Sapiens
2000s
Kofax · ILOG
2010s
Lexmark · Solaris
2020s
Haystac
Barak Tsivkin
CEO & Co-Founder
20+ years in enterprise software. Founded Solaris Development (Partners Healthcare, MGH, Dana Farber). Early sales at ILOG — helped scale to $30M in five years.
Eli Zukovsky
CTO & Co-Founder
30+ years in distributed systems, rules engines, and computational linguistics. Co-founder and principal architect at Sapiens. M.S., Latvian State University.
Vadim Ivanov, Ph.D.
Chief Scientist
20+ years in ML research. Author of the core algorithms behind Haystac — scalable classification, conceptual clustering, fuzzy pattern matching for noisy text.
Anthony Macciola
Chairman of the Board
CIO at ABBYY. Former CTO at Kofax — led the company’s move into mobile capture, NLP, and process automation. 45+ patents in text analytics and image processing.
100+
Combined years building enterprise content platforms
170+
Patents across the team — 45+ from Anthony alone
3
Generational platform transitions navigated
Haystac applies that playbook to the next generation of AI — with a relentless focus on control, domain-specific intelligence, and real-world adoption in regulated industries.
Gartner 2026: Domain-Specific Language Models named a Top Strategic Technology Trend. Gartner projects the DSLM market will reach $131B by 2035.
Deployment
Whatever your security review demands, we have a path.
Haystac goes to market through partners — VARs, OEMs, and BPOs already operating inside regulated environments. They handle the deployment. We handle the platform.
Most regulated
On-premises
Maximum control. Your hardware, your network, your air gap if you need one.
InfraYour hardware
DataNever leaves your perimeter
GPUOn customer hardware
Live inWeeks (on existing infra)
Best for
Federal · Defense · Top-tier banks
Fastest to start
Private cloud
Skip the hardware procurement. Deploy into your own AWS, Azure, or GCP account.
InfraCustomer-managed VPC (AWS · Azure · GCP)
DataStays in your cloud account
GPUOn demand
Live inDays
Best for
Insurers · Health systems · Regional banks
Best of both
Hybrid
Sensitive ingest stays on-prem. Reasoning runs in private cloud. One platform across both.
InfraMix of on-prem & cloud
DataSensitive content on-prem
GPUOn demand for reasoning
Live inWeeks
Best for
Mature enterprises in transition
FAQ
Common questions from security, ops, and IT.
Can it really run entirely inside our environment?
Yes. Fully containerized (Docker). On-prem, customer-managed VPC, or hybrid. No outbound connectivity. No SaaS dependencies. No telemetry phoning home.
How does it handle hallucinations?
RAG and Graph RAG architecture. Every response is constrained to retrieved enterprise content, with line-level source attribution. The model can’t answer from outside your corpus — that’s the point.
How long does deployment take?
Day One inside existing infrastructure. DIY private AI takes 12–24 months. We measure deployment in weeks, not quarters.
How does it integrate with our existing systems?
REST API. MCP for tool invocation. Runs alongside your ECM, CRM, ERP, and case systems. No rip-and-replace.
How are you priced?
Partner-led. Talk to a Haystac partner for a deployment quote scoped to your environment, document volume, and use cases.
Ready when you are
See what AI looks like inside your environment.
Bring your documents, your systems, and your compliance constraints. We’ll show you how Haystac works inside the environment you already control — in 30 minutes.
Request a demo →