Custom emotion recognition software, built on a proven engine
Most custom emotion detection projects start by trying to build a model. We start with one that already works. EchoDepth reads emotion in video, voice, text and images using the Facial Action Coding System - and we build your product, workflow and reports on top of it, deployed where your data needs to stay.
What you get on day one
A scientific standard
FACS-based facial analysis across 44 Action Units, calibrated across 14 cultural and linguistic cohorts - not a generic image classifier.
Every modality
Endpoints for video, audio, text, image and multimodal analysis, plus WebSocket streaming for real-time video.
Developer tooling
REST API, Python and Node.js SDKs, and webhooks that post full results when each analysis completes.
Your brand or ours
White-label reports for agencies and consultancies, or your own interface built on the API.
Data where it must stay
UK by default, EU-sovereign or fully on-premise. ICO registered ZB915633, with DPIA support.
Scores you can defend
Documented methodology and confidence intervals, so results stand up to procurement and regulators.
Already built on the same engine
We have used EchoDepth to build products for very different markets. Yours would start from the same foundation.
How a build works
- 01Scope
Your use case, your users, and the law where you operate - including the EU AI Act and a DPIA - before any build starts.
- 02Pilot
Run EchoDepth on your own content to prove the signal is there. Pilots start from £1,500 per month.
- 03Build
Integrate by API, SDK and webhook, or have us build the interface and reports around your workflow.
- 04Deploy
UK, EU-sovereign or on-premise, under your brand or ours, with the methodology documented.
Build from scratch, or build on EchoDepth?
| Build a model from scratch | Build on EchoDepth | |
|---|---|---|
| Starting point | Data collection and labelling | A working engine across four modalities |
| Scientific basis | Whatever the training data teaches it | FACS, 44 Action Units |
| Cultural bias | Usually trained on one population | Calibrated across 14 cultural and linguistic cohorts |
| Governance | Written from nothing | ICO registered, DPIA support, documented methodology |
| Where effort goes | The model | Your product, workflow and users |
Scoped against the law first
Emotion recognition is regulated differently in every market - the EU AI Act prohibits it in workplaces and education. Every build starts with where it will be used and who it reads. See our guides to the EU AI Act, UK GDPR and the law by jurisdiction.
Questions
Can Cavefish build custom emotion recognition software for us?
Yes. Cavefish builds custom emotion recognition and emotion detection products on its own engine, EchoDepth, rather than starting from scratch. You get a working analysis engine on day one - video, voice, text and image - and the build effort goes into your workflow, your interface and your data rules. The same engine already runs Cavefish's own products for defence, sales, research, sport and fintech.
Why build on EchoDepth instead of training our own emotion detection model?
Training a defensible emotion model needs a scientific standard, annotated data across cultures, validation and governance before it produces anything useful. EchoDepth is built on the Facial Action Coding System (44 Action Units) and calibrated across 14 cultural and linguistic cohorts, with documented methodology and confidence intervals. Building on it moves you from research project to product, and leaves you with evidence you can defend to a regulator or procurement team.
How do we integrate EchoDepth into our own product?
Through the EchoDepth REST API (video, audio, text, image and multimodal endpoints), WebSocket streaming for real-time video, webhooks that post full results when an analysis completes, and Python and Node.js SDKs. Results can be pushed into tools such as Salesforce, HubSpot and Microsoft Teams. For agencies and consultancies, reports can be delivered white-label under your brand.
Where is the data processed?
In the UK by default, with EU-sovereign and on-premise deployment for regulated or data-sensitive environments, where data never leaves your infrastructure. Cavefish is registered with the ICO (ZB915633), supports your Data Protection Impact Assessment, and works under UK GDPR.
Is custom emotion recognition legal?
It depends on where and how it is used. The EU AI Act prohibits emotion recognition in workplaces and education (with narrow medical and safety exceptions) and treats most other uses as high-risk. EchoDepth is designed to analyse content - a pitch, an advert, a message, a call - rather than to monitor people, and we scope every build against the law in the countries you operate in before any code is written.
Tell us what you want to build
A 30-minute conversation with the team about your use case, your data and where it needs to run.