Voice AI Real-Time Executive Voice AI via Semantic Edge Caching
- Coordinator
- Intelswift Ou, Estonia
- Runs
- 1 September 2026 to 31 August 2028
- Status
- Running
- EU funding
- €500k
- Total cost
- —
- Partners
- 1 organisations in 1 countries
- Type
- European Innovation Council
- Grant agreement
- 101310310
What the project is doing
In the project's own words, from its summary on CORDIS.
The Problem automated customer service market is in crisis. Traditional IVR and first-gen AI bots are failing due to three systemic flaws:Unnatural Latency: Average response times of 2.8s+ destroy conversation flow (human threshold is <1s). Passive Functionality: Bots only ""talk"" (provide info) but cannot ""do"" (execute transactions like blocking a card). The ""Language Gap"": Big Tech focuses on English, leaving European markets (especially CEE: UA, PL, RO) with poor recognition accuracy and robotic voices.
The Solution Intelswift is engineering Agentic Voice AI, a transition from informational bots to executive agents that act with human speed. We are advancing this Deep Tech from TRL 4 to TRL 7 through four proprietary innovations:Semantic Edge Caching (Speed): We solve the latency problem not by optimizing LLMs, but by building a proprietary middleware layer. This architecture intelligently caches responses at the network ""edge,"" radically slashing response times to ≤1.5 seconds to match human perception.
Secure AI Actions (Execution): We implement a secure ""Function Calling"" framework with 2FA, enabling agents to autonomously execute complex transactions (e.g., changing delivery addresses or updating CRM data) in real-time, rather than just retrieving answers. Noise-Adaptive VAD (Robustness): To solve reliability issues, we are developing a hybrid Voice Activity Detection model with a noise-adaptive transformer layer, ensuring >85% accuracy even in noisy environments (>65dB).
Strategic Localisation (Quality): Rather than simply ""supporting"" languages, we are solving the quality deficit. We are building a ""Unified Voice Profile API"" to deliver high-fidelity, natural voice models for 24 EU languages, specifically engineering superior accuracy for the under-resourced markets (Poland, Romania, Ukraine) that global competitors currently fail to serve effectively.""
Who is involved
| Organisation | City | Country | Role |
|---|---|---|---|
| Intelswift Ou | Tallinn | Estonia | Coordinator |
Source: CORDIS, European Commission. Data of 5 Nov 2026, reused under the Commission's reuse policy. View on CORDIS · Grant DOI