The SentraLink Engine
LLM-powered extraction grounded in source evidence. Graph-based correlation. Format-agnostic ingestion.
Core Intelligence Engine
Every insight is traceable back to the source evidence that produced it
Zero-Config Cross-Modal Chaining
SentraLink's core differentiator: it chains evidence across modalities automatically. A chat mention links to a photo, a face in that photo to another image, a voice in a call to a second recording — multi-hop chains built on ingestion. No analyst defines these links by hand, and no ontology setup is required.
LLM Extraction with Evidence Grounding
Large language models extract entities, relationships, and intent from unstructured evidence. Every extraction is grounded — linked back to the exact source passage, page, or record it came from. No hallucinated connections. Investigators can verify any finding against the original material.
Graph-Based Correlation
Extracted entities are mapped into a knowledge graph that reveals hidden relationships across evidence types. Temporal matching links communications to transactions. Community detection algorithms surface organizational structures, key players, and coordination patterns that linear review cannot find.
Face Clustering & Voice Signatures
Vision AI clusters the same face across all case photos; speaker biometrics match the same voice across separate recordings. Both link visual and audio evidence into the correlation graph. These biometric modules are optional and configurable per jurisdiction.
Semantic Search Across Case Data
Search by meaning, not just keywords. Ask questions in natural language and retrieve relevant passages across CDRs, transcripts, financial records, and documents simultaneously. The search engine understands investigative context — "who received money after the call" returns results across evidence types.
Format-Agnostic Ingestion
Upload evidence as it arrives: device extractions from Cellebrite, MSAB XRY, Magnet AXIOM and Oxygen, disk images, mailboxes, CDRs, bank statements, warrant returns, scanned documents, photos, video and recorded calls. Extraction packages are unpacked automatically and spreadsheets are classified on sight, so no conversion to loose files and no format-specific setup is needed.
See It in the Product
Screens from demonstration cases built on synthetic and reconstructed evidence
One graph for the whole case
Every call, document, photo, person, place and amount sits in one graph, with a timeline of the case across the top. The selected recorded call shows its risk reason and summary beside the graph, and each link can be opened to see which two files it joins.
Risk with a written reason, on the map
Each item is scored for risk, and the reason is written out. Here a call is flagged because the speakers arrange a cash payment for a building permit. The places named in the call are pinned on the map together with the locations from the carrier records.
Ask the case, get the source
The case opens with an automatic briefing. Questions in plain language are answered only from the evidence in the case, and every answer names the recording or document it came from.
Weapons and narcotics in photos
Photos are described in full and objects are detected automatically. This seized photo is flagged for two weapons and a drug item, then linked to the conversations about the same shipment.
Visual similarity across unrelated photos
Images are matched on parts of the picture as well as the main subject. These two photos were linked on the mountain range in the background, which places them in the same area.
Evidence in its original language
The engine understands evidence spoken or written in different languages and correlates between them regardless of language. Patterns such as payments and envelopes are found across independent sources even when the words differ.
Intelligence Capabilities
From raw evidence to actionable investigative picture
Conversational Case Intelligence
Chat with your case data in natural language. Ask questions, explore connections, and receive context-aware answers drawn from correlations across all ingested evidence. Set Case Directives to tell the AI what matters — suspect names, transaction patterns, specific leads — and every analysis is shaped by those priorities.
Image & Object Correlation
Vision AI detects objects, scenes, and patterns in photos — weapons, vehicles, locations, logos, documents — not just faces. Each visual finding is matched across every image in the case and chained to where it appears in other evidence: a vehicle in a seized photo linked to the same vehicle described in a wiretapped call, named in a document, or captured in another image. Recurring patterns and anomalies across the visual evidence surface automatically.
Over 100 Languages, Slang & Code Words
Speech and text in over 100 languages are read in the original script, including items that mix languages, with automatic detection and no pre-configuration. Summaries, entities and correlations are presented in English, and the case chat accepts questions in any language about material in any language. Extraction models are trained on real-world criminal communication patterns, including coded language, street slang and obfuscation.
Temporal Pattern Discovery
Automatically discovers meaningful time relationships between events across data sources — linking communications to financial actions, detecting coordinated patterns, and revealing cause-and-effect sequences investigators would otherwise miss.
Entity Resolution Across Sources
The same person appears differently across phone records, bank statements, and chat exports. SentraLink resolves entities across formats — matching phone numbers, names, aliases, and account identifiers into unified profiles automatically.
Financial Flow Mapping
Track money movement across banking systems and cryptocurrency ledgers. Follow funds through layering operations, identify beneficial owners behind shell companies, and flag mule accounts — all correlated with communications data.
Geospatial & Location Correlation
Resolve location from cell-site and CDR data, GPS logs, ALPR reads, and photo geotags into a single movement picture — correlating where a subject was with what they said and did, and surfacing pattern-of-life and co-location across the case.
Anomaly & Pattern Detection
Surface the outliers — an unusual transaction, an off-pattern contact, a break in routine — by establishing what is normal across each subject's communications, finances, and movements, then flagging what does not fit for analyst review.
Video Analysis
Video is analysed frame by frame. Faces, vehicles, licence plates, logos and other objects are detected and tracked, and a clip is extracted for each one. Every clip is correlated with the photos, calls and documents that show or mention the same face, plate or object, so a vehicle seen in seized footage links to the call that describes it.
Visual Similarity
Images are compared visually, not only by their written description. Matching works on parts of a photograph as well as its subject, so two unrelated photos that share the same mountain range, interior or distinctive object in the background are linked, giving the investigator a basis to place them in the same location.
Risk Triage & Leads
Every item receives a risk assessment with a written reason, and everyday noise such as one-time codes, broadcasts and marketing is demoted. When a routine item correlates with a high-risk one, its risk rises with it. The investigator starts from a ranked list of leads and corroborated stories rather than from the file list.
Court-Ready Evidence Packaging
Findings are correlated to source data with full traceability. Chain of custody protection, cryptographic audit logs, and exhibit authenticity certification ensure judicial-grade reports meet evidentiary standards.
Supported Evidence
Several extraction formats can be loaded into the same case and correlated with each other
Device Extractions
Cellebrite (UFDR / UFED), MSAB XRY, Magnet AXIOM, Oxygen, E01 and raw disk images. Unpacked automatically into their messages, calls, contacts, media and files. Vendor signatures are validated on ingestion where present.
Recorded Audio
WAV, MP3, M4A, FLAC, OGG, WMA, AAC. Transcription in the original language, speaker separation and voice signatures matched across calls.
Images & Video
JPG, PNG, BMP, TIFF, HEIC and common video formats. Scene description, object and weapon detection, face clustering, licence plates, text in images, EXIF and geolocation.
Email & Documents
PST, OST, MBOX, EML, EMLX and MSG with attachments. PDF including scanned pages with OCR, Word, Excel, PowerPoint, RTF and text.
Structured Records
CSV, Excel and JSON. Bank statements, carrier call detail records, ledgers, registers and transaction exports are recognised and classified on ingestion.
Warrant Returns & Chats
Google Takeout, Facebook warrant returns and other provider returns, plus chat exports and in-device conversations recovered from extraction packages.
Built for Law Enforcement Infrastructure
Secure, compliant, and integrated with existing systems
On-Premise & Air-Gapped
Runs entirely within your infrastructure. No cloud dependencies. All AI models run locally — no external API calls. Compatible with air-gapped and classified environments.
Enterprise Integration
Single sign-on with existing authentication. Active Directory support. Direct file system access. Works with your current case management systems.
GPU-Accelerated Processing
NVIDIA GPU support for LLM inference and graph processing. Process large datasets faster. Real-time entity extraction from communications and documents at scale.