Intent-aware Multi-lingual AI Agents for Voice-Based Transactions in Edge IoT Systems
Bibliographic record
Abstract
Voice-based commerce is emerging as a convenient modality for consumers to transact in domains like retail and automotive services. However, deploying intent recognition agents in noisy, bandwidth-limited Internet of Things (IoT) environments (e.g., vehicles, factories) poses significant challenges. This paper proposes an intent-aware AI agent framework optimized for on-device speech recognition and understanding in edge IoT settings. The system leverages robust acoustic modeling and multilingual spoken language understanding to reliably extract user intents under ambient noise and network constraints. We introduce an edge-compute architecture that processes voice commands locally for low latency and privacy, using model compression and quantization to fit resource-constrained devices. To enable monetization, the agent supports contextual product recommendations (voice-based “ads”), third-party API call integrations with licensing control, and a tiered service model where premium subscribers benefit from enhanced on-device inference capabilities. We evaluate the framework in a smart automotive environment. The on-device system achieves 94.3% intent accuracy in quiet conditions and$\mathbf{9 0. 5 \%}$under noisy conditions with noise suppression (SNR$\approx \mathbf{5 - 1 0 ~ d B}$), with 280 ms end-to-end latency and a quantized intent model of 0.30 MB (-75% vs. uncompressed). These results are comparable to cloud-based performance while reducing latency and preserving privacy.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".