Platform Capabilities
How MergeOn Actually Works.
The Technical Deep Dive.
No marketing fluff. Here's exactly how we achieve zero hallucination, sub-second ingestion, model agnosticism, and cross-document intelligence.
Speed
Sub-Second Ingestion
Most document AI platforms take minutes per page. We process documents in milliseconds. Here's how:
Parallel Field Extraction
Every field extracted simultaneously, not sequentially. 100 fields = same time as 1 field.
Pre-compiled Templates
Document schemas are pre-indexed. Recognition is O(1) lookup, not O(n) search.
Streaming Pipeline
AI detection, OCR, and verification stages run in parallel streams, not blocking queues.
Edge Caching
Frequently-seen document types cached at edge. Repeat documents are instant.
Processing Time Comparison
Traditional OCR45s
GPT-4 Vision12s
MergeOn0.3s
150xfaster than traditional methods
Evidence-Linked Architecture
SOURCE DOCUMENT
Page 3, Coords [142, 287, 445, 312]
EXTRACTED FIELD
purchase_price: $1,250,000
SHA-256 HASH
e3b0c44298fc1c149afbf4c8...
AI OUTPUT
"The purchase price is $1.25M"
↑ Linked to evidence above
Truth
Zero Hallucination by Design
LLMs hallucinate because they generate from patterns, not facts. We eliminate this structurally:
Every Field Has Coordinates
Not just the value — the exact pixel location on the source document.
SHA-256 Evidence Hashing
Cryptographic proof linking every output to its source. Tamper-evident.
Human Verification Gate
AI detects, humans verify at 100% accuracy. No unverified claims reach output.
Confidence Thresholds
Conservative AI scores force human review. We'd rather flag than guess.
"If you can't point to the pixel, you can't claim the fact."
Architecture
True Model Agnosticism
Most "model-agnostic" claims are marketing. Here's how we actually achieve it:
Truth Lives Outside the Model
Your knowledge graph is stored independently. Models read from it, never own it.
Standardized Context Packets
MIL formats context identically for any LLM. Same input structure, any provider.
No Fine-Tuning Required
We don't embed knowledge in weights. Switch models without losing anything.
Provider Failover
If OpenAI is down, route to Claude. If Claude is slow, route to Gemini. Automatic.
Hot-Swap Any Model
YOUR KNOWLEDGE
DealMap + Evidence Graph
GPT-4
Claude
Gemini
Llama
Switch providers in config, not code. Zero re-processing.
Intelligence
Cross-Document Field Dependencies
Real transactions span dozens of documents. A change in one affects obligations in others. We track every dependency.
Purchase Agreement
closing_date
March 15, 2025
Loan Commitment
funding_deadline
closing_date - 3 days
← References Purchase Agreement
Title Insurance
effective_date
closing_date
← References Purchase Agreement
⚡ Automatic Cascade Detection
If closing_date changes in the Purchase Agreement, MergeOn automatically flags the Loan Commitment and Title Insurance as affected, recalculates derived dates, and alerts relevant parties.
Dependency Graph
Every cross-reference tracked as a directed edge. Full lineage visibility.
Change Propagation
Modify one field, see every downstream impact instantly.
Conflict Detection
Spot contradictions across documents before they become legal problems.
Missing Document Discovery
AI detects when referenced documents are absent from the corpus.
Accessibility
The Blue Dot & Mouse Tracer
MIL doesn't just protect data — it understands context. The blue dot and mouse tracer create a universal accessibility layer:
◉
Attention Detection
Tracks where users hover, pause, and struggle. Builds a real-time understanding of confusion points.
◎
Contextual Assistance
When someone hovers on a field for 2+ seconds, MIL knows they might need help — without them asking.
◇
Adaptive Explanations
A child gets simple language. An elderly user gets larger text and step-by-step guidance. A lawyer gets technical precision.
◆
Form Completion Aid
Elderly users filling out medical forms get voice-guided assistance. Kids doing homework get age-appropriate hints.
Live Context Understanding
Patient Name
Insurance Policy Number
"I see you're on the policy number. It's usually on your insurance card, top right corner."
Date of Birth
MIL detects struggle → Provides contextual help → No PII sent to LLM
More Capabilities
Additional Technical Features
Coordinate Refinement
Multi-pass boundary refinement achieves sub-pixel accuracy. Every field extraction includes exact coordinate mapping.
Template Learning
Process 3 examples of a form type, MergeOn learns the schema. New documents auto-classify.
Version Diffing
Upload Amendment 3 — instantly see what changed from Amendment 2, field by field.
Audit Logging
Every extraction, verification, and query logged with timestamps. Complete forensic trail.
Webhook Events
Real-time notifications when documents complete processing, fields change, or conflicts detected.
Batch Processing
Upload 10,000 documents. Get 10,000 structured outputs. Linear scaling, parallel processing.
See It In Action
These aren't slides — they're live capabilities. Request a technical demo and we'll show you the actual system.
Request Technical Demo