Heimdall SIGINT-I
Illuminating the Information Battlespace
Strategic-grade detection at scale. The full Heimdall pipeline — 56 span-level techniques, 8 stealth analyzers, 16 flow patterns, 7 emotion classes, 7 hierarchical categories — with dynamic batching (up to 32 requests per GPU pass), LRU caching (10K entries, 168-hour TTL), and LLM-augmented deep analysis via Ollama, OpenAI, Gemini, or Claude. Process thousands of articles per hour with sub-second latency per analysis.
Challenges You Face
Scale of Analysis
Dynamic batching processes up to 32 requests per GPU forward pass with 100ms collection window — but need analyst workflows on top
Attribution Challenges
Need technique fingerprinting, stealth pattern correlation, and source balance analysis across dozens of outlets to attribute campaigns
Early Warning
Detect coordinated messaging, technique accumulation patterns, and fear escalation before campaigns mature
Cross-Platform Correlation
Same techniques, same stealth patterns, same arc shapes appearing across platforms — need to connect them
Stealth IO at Scale
State actors using 8 stealth techniques across "independent" outlets — need to detect coordination in stealth patterns
LLM-Augmented Analysis
Analysts need deeper context per detection — explanations, counter-arguments, and narrative analysis at scale
Key Features
Full Pipeline at Scale
POST /analyze/full runs propaganda + stealth + flow + entities + keywords + sentiment + moderation in parallel via asyncio.gather
Dynamic Batching
Up to 32 requests queued per GPU forward pass with 100ms collection timeout — single pass resolves all futures
LRU Cache
10,000 entries with 168-hour TTL — repeated analysis of same content returns instantly
LLM Explanations Pipeline
Per-technique explanations with counter-arguments via Ollama (llama3.1:8b-instruct-q8_0), OpenAI (gpt-4o-mini), Gemini, or Claude — with fallback chain
7 Hierarchical Categories
EMOTIONAL_MANIPULATION, SOCIAL_PROOF, LOGICAL_FALLACIES, SOURCE_MANIPULATION, IDENTITY_ATTACKS, REPETITION_AND_SLOGANS, INFORMATION_MANIPULATION
State Actor Fingerprinting
Technique distribution + stealth pattern + arc shape correlation across outlets for campaign attribution
Batch Analysis API
POST /analyze/batch processes multiple texts per call — each with full 56-technique + severity + emotion output
Indicator Export
Structured JSON output compatible with STIX/TAXII threat intelligence sharing frameworks
What Others Miss
Stealth manipulation that hides behind objectivity. Heimdall exposes it.
Track perspective_ratio across dozens of outlets over time — identify systematic sourcing bias as an IO indicator
SpaCy NER + VADER valence per entity across outlet networks — detect coordinated framing patterns
Systematic omission of unfavorable context across multiple outlets as a coordination indicator
Same doubt-planting phrases appearing across "independent" outlets — 40 patterns matched and correlated
Track active/passive voice ratio per entity across outlets — coordinated agency obscuration as IO fingerprint
Techniques We Detect
State-level synchronized operations — technique distribution fingerprinting across outlets
Manufactured movements at scale — coordinated inauthentic behavior detection
Overwhelming information space — technique accumulation (ESC003) flow pattern detection
High-volume, multi-channel manipulation — intelligence flow pattern with rapid technique variation
Pricing
Team
- ✓Collaborative analysis
- ✓Batch API
- ✓Full stealth suite
- ✓Campaign correlation
Enterprise
- ✓Full deployment
- ✓Custom model fine-tuning
- ✓LLM pipeline
- ✓Priority GPU allocation
Full Pipeline at Scale. Sub-Second Latency.
Schedule a demo or start your free trial today.