What Is Raven? Graylark’s AI Investigation Platform
Investigations often start with fragments. Raven helps investigators turn a photograph, an email address, or a phone number into reviewable leads.
Raven is an AI investigation platform developed by Graylark. Law-enforcement agencies, government teams, and other investigative organizations use it to estimate where an image was taken, discover public accounts associated with an email address or phone number, identify vehicles, check images for signs of manipulation, and organize their findings in one case workspace.
Raven returns ranked leads, not verdicts. The investigator reviews each one, corroborates it, and decides what it means.
What is Raven?
Investigations rarely begin with a complete picture. They begin with fragments: a photograph shared online, a frame pulled from a video, an email address in a complaint, or a phone number on a listing.
Each fragment can hold a lead. A photograph can show where it was taken even after its metadata, the GPS and camera details some files carry, has been stripped. An email address can be linked to public accounts that reveal a username, a review, or another image.
Working those fragments by hand means separate tools, copied screenshots, and notes kept elsewhere. Raven brings the work into one platform. Computer-vision models, AI trained to interpret images, analyze each input and return ranked candidates. The investigator reviews them and connects what holds up to the rest of the case.
What can Raven do?
Each capability takes a specific input and returns candidates for review. Similarity scores order candidates against one another. They are not probabilities that a result is correct.
Find Region
Start with a photograph or video frame, even one with no landmark or metadata. Find Region returns ranked candidate areas, each with coordinates and an approximate radius. This is image geolocation: estimating where a picture was taken from visible terrain, vegetation, architecture, and road design. Use it to decide where to look first.
Find Street
Select a region, from Find Region or case information, and Find Street, also called street targeting, ranks candidates inside it with meter-level accuracy: a specific street, building, or address, not a neighborhood. Each has an address and coordinates to compare with reference imagery, meaning street-level photographs of that location. Results depend on the image and the region’s coverage.
Find Property (Beta)
Choose a supported city and upload an interior or exterior photograph. Find Property ranks candidate properties and returns the photographs behind each match for side-by-side comparison. It is a beta capability in selected cities and searches one city at a time. Coverage and performance vary by city.
Live Intelligence
Start with an email address or phone number. Live Intelligence checks supported online services and returns publicly available accounts, usernames, reviews, and images, each with its source and date. This is open-source intelligence (OSINT): information drawn from publicly accessible sources.
“Live” means the check runs when the search is submitted. It is not continuous tracking, and it does not log into accounts or reach private messages. A mapped review marks the place reviewed, not where someone is now. An account association is a lead to verify, not proof of ownership. Any returned image can go straight into Find Region and Find Street.
Identify Car
Upload an interior or exterior view of a vehicle, even a partial, obstructed, or blurred one. Identify Car returns ranked make, model, and year candidates with reference images to compare. It reads the vehicle itself, not the license plate.
Verify Image (Beta)
Verify Image assesses signals that an image was generated by AI or altered through a deepfake or face swap. It returns a plain-language verdict, the generator the signals most resemble, and a separate face-manipulation reading. It is a beta preview, and accuracy is not guaranteed. Treat the result as a reason for closer scrutiny, not a determination of authenticity.
Cases
Cases is where an investigation comes together. Source imagery, search results, locations, investigator pins, and notes sit on one map. Each candidate keeps the review state the investigator recorded: confirmed, possible, or dismissed.
How does Raven fit into an investigation?
This example is hypothetical. It illustrates the workflow, not a real case.
An investigator finds an online listing for a car priced far below market value. There is no address, and the seller asks buyers to move to private messages. The photograph is the best evidence available.
The investigator runs it through Verify Image to check for signs of AI generation, and through Identify Car for vehicle candidates to compare with the listing’s description.
Then the investigator turns to the background: the houses, driveway, and landscaping behind the car. Find Region returns ranked candidate areas, and Find Street searches the leading one for street-level candidates.
Each candidate opens beside reference imagery. The investigator compares rooflines, garage doors, and neighboring buildings, dismissing most. One agrees on several independent features and is marked possible, pending corroboration outside Raven.
Finally, the investigator adds the searches to a case. The photograph, the vehicle and location candidates, and the notes sit in one workspace, where another reviewer can follow how the lead was developed. If the listing includes a phone number, a Live Intelligence lookup can add associated public accounts to the same case.
None of this proves the car was stolen or identifies the seller. It produces a reviewed lead to develop through normal procedure. Geolocating a Suspected Stolen Car From a Single Social Media Post shows a similar workflow, recorded in Raven.
How are Raven, GeoSpy, and Graylark connected?
Graylark Technologies is the company, an applied AI research lab. Raven is its investigation platform, which Graylark describes as frontline visual intelligence.
Raven grew out of GeoSpy, Graylark’s earlier AI image geolocation system. Investigators using GeoSpy needed more than a regional estimate: street-level refinement, vehicle identification, image verification, and a place to organize results. In April 2026, GeoSpy became Raven to reflect that broader platform, and its geolocation capabilities continue inside Raven.
GeoSpy is now Raven explains the transition, and Why We Built GeoSpy and Why It Became Raven tells the origin story.
Who is Raven for?
Raven is built for professionals who develop leads from material with little context: detectives, analysts, and task-force investigators in law enforcement; analysts and investigative teams in government, public-safety, and intelligence organizations; and corporate investigators, fraud teams, and security analysts in enterprise settings.
Raven is available to verified agencies and investigative teams, after review. Evaluation usually starts with a demonstration and a discussion of your requirements. Trials may be available to qualified organizations, and Graylark may require onboarding before evaluation access.
Security, privacy, and procurement documentation can be requested during an evaluation. Customer imagery submitted to Raven is not used to train Raven’s models. The Raven FAQ covers reading results, data handling, and purchasing.
Getting started with Raven
The clearest way to judge Raven is to see it run on the kind of material your team handles: a difficult photograph, a frame from a video, or an identifier from a case file. Book a demonstration using representative inputs, or explore Raven’s capabilities first.
See Raven work on your own inputs.
Raven is available to verified agencies and investigative teams. Book a demonstration using representative imagery, email addresses, or phone numbers from your work.
