Meaningful work and stability, but slow execution and pay issues
Pros
1. Meaningful impact (this one is real) You’re not optimizing ad clicks—you’re directly improving quality of life. Hearing loss is deeply tied to cognitive decline, social isolation, and mental health Even small model improvements (noise suppression, speech enhancement, directionality) can have immediate human impact You actually see your work in a physical product used daily 2. End-to-end ownership The “you do everything” downside is also an upside: You touch research → model → embedded deployment → product 3. Stability & low volatility Not a hype-driven startup Less risk of layoffs compared to big tech cycles 4. Unique domain expertise You build skills that are: Hard to replicate elsewhere Valuable in: medical devices edge AI audio ML
Cons
1. “Old tech” is often accurate Legacy codebases (often C/C++ + DSP pipelines) Slow adoption of: modern deep learning stacks cloud-native infra Tooling can feel years behind companies like Google or Meta 2. Very slow execution speed Medical/device constraints + corporate structure = long cycles Shipping can take months (or longer) Many approvals, validations, compliance steps 3. Too many management layers Decision-making often: top-down slow sometimes disconnected from engineering reality 4. Lack of MLOps / DevOps culture This is a big one for ML engineers: No mature: CI/CD for models experiment tracking infra scalable pipelines You’re often expected to: train models deploy them optimize embedded inference manage infra yourself 5. Limited career growth (in many cases) Fewer: senior technical ladders cutting-edge ML teams Promotions can be: slow tenure-based rather than impact-based 6. Compensation gap Typically below: FAANG top-tier startups Especially noticeable for: senior ML engineers research-oriented roles