Problem
The product needed to combine vehicle information and conversational controls on hardware with strict limits on memory, compute, latency and connectivity.
03 / CASE STUDY
Embedded AI · Automotive voice system
S61 Car Assistant
An automotive assistant combining real-time diagnostics, voice interaction and predictive-maintenance workflows on resource-constrained hardware.
RoleAI systems engineer · Part-time
StatusClient automotive system
Timeline2022 — 2025 engagement
01 / CONTEXTThe problem
The problem
behind the product.
Approach
I treated inference cost as an architectural constraint, optimizing the language and speech pipeline around the Raspberry Pi environment and separating real-time diagnostics from conversational tasks.
02 / SYSTEMArchitecture,
Architecture,
not magic.
01Vehicle signals→
02Diagnostics layer→
03Embedded NLP→
04Decision logic→
05Voice interface
Constraints
- Resource-constrained hardware
- Real-time response expectations
- Automotive integration
- Intermittent connectivity
My ownership
- Built and optimized NLP pipelines
- Deployed Transformer-based components to Raspberry Pi
- Integrated diagnostics with voice workflows
- Troubleshot performance across hardware and inference layers
03 / EVIDENCEWhat can be
What can be
defended.
Embedded Transformer deployment
Voice-controlled vehicle workflows
Real-time diagnostic integration
Predictive-maintenance capability
What I would improve now.
The enduring lesson was that embedded AI succeeds through careful system partitioning and latency budgets, not simply by selecting a smaller model.
PythonTransformersRaspberry PiSpeech AIEmbedded Linux