Sarvam
About this role
This role owns Sarvam's entire Intel surface end to end: Intel NPU through OpenVINO on Meteor Lake, Lunar Lake and vPro AI PCs, Intel integrated and discrete GPU through OpenVINO and ONNX Runtime, and x86 CPU optimisation for the fallback path. You would land Sarvam's edge models on Intel NPU, discrete GPU and integrated GPU inside defined service level agreements, own the OpenVINO build and quantisation recipe for the team's models including the driver version compatibility matrix, and drive x86 CPU optimisation using AVX-512, AMX and threading strategy. You also own the Intel device CI pool and regression detection across OpenVINO upgrades. The posting describes you as the technical face to Intel's ecosystem and the OpenVINO team, so this is a vendor facing engineering seat as much as an internal one. It is a narrow specialism: 5+ years on ML deployment with 2+ specifically on Intel inference stacks.
Who this is for
What the posting requires:
- 5+ years on ML deployment, with 2+ years specifically on Intel inference stacks.
- Production OpenVINO experience, including model conversion, accuracy validation after quantisation, and driver version pinning.
- ONNX Runtime execution provider knowledge: when to use the OpenVINO EP versus native OpenVINO, and when to fall back to the CPU EP.
- x86 CPU profiling and optimisation with VTune and perf, including comfort reading hot loops at the assembly level when needed.
Noted as a strong plus rather than a requirement:
- AVX-512 or AMX intrinsics.
Bonus points:
- Direct prior interaction with the OpenVINO team or Intel ecosystem partners.
- Custom OpenVINO operator authoring.
What the work actually looks like:
- Land Sarvam's edge models on Intel NPU, discrete GPU and integrated GPU inside defined service level agreements.
- Own the OpenVINO build and quantisation recipe for the team's models, including the driver version compatibility matrix.
- Drive x86 CPU optimisation for the universal fallback path, covering AVX-512, AMX and threading strategy.
- Own the Intel device CI pool and regression detection across OpenVINO upgrades.
The hardware surface named in the posting: Intel NPU via OpenVINO on Meteor Lake, Lunar Lake and vPro AI PCs, Intel integrated and discrete GPU via OpenVINO and ONNX Runtime, and x86 and AMD64 CPU for fallback paths.
Location and working pattern: Bengaluru. No remote or hybrid arrangement is stated.
Honest fit guidance: the 2 years on Intel stacks specifically is the real filter, not the 5 years overall. General ML engineering or even CUDA and Nvidia deployment experience does not substitute, because the entire job is the Intel toolchain and its quirks. If you have that background it is a rare seat, since few teams in India run this surface at production scale.
- 5+ years on ML deployment, with 2+ years specifically on Intel inference stacks.
- Production OpenVINO experience, including model conversion, accuracy validation after quantisation, and driver version pinning.
- ONNX Runtime execution provider knowledge: when to use the OpenVINO EP versus native OpenVINO, and when to fall back to the CPU EP.
- x86 CPU profiling and optimisation with VTune and perf, including comfort reading hot loops at the assembly level when needed.
Noted as a strong plus rather than a requirement:
- AVX-512 or AMX intrinsics.
Bonus points:
- Direct prior interaction with the OpenVINO team or Intel ecosystem partners.
- Custom OpenVINO operator authoring.
What the work actually looks like:
- Land Sarvam's edge models on Intel NPU, discrete GPU and integrated GPU inside defined service level agreements.
- Own the OpenVINO build and quantisation recipe for the team's models, including the driver version compatibility matrix.
- Drive x86 CPU optimisation for the universal fallback path, covering AVX-512, AMX and threading strategy.
- Own the Intel device CI pool and regression detection across OpenVINO upgrades.
The hardware surface named in the posting: Intel NPU via OpenVINO on Meteor Lake, Lunar Lake and vPro AI PCs, Intel integrated and discrete GPU via OpenVINO and ONNX Runtime, and x86 and AMD64 CPU for fallback paths.
Location and working pattern: Bengaluru. No remote or hybrid arrangement is stated.
Honest fit guidance: the 2 years on Intel stacks specifically is the real filter, not the 5 years overall. General ML engineering or even CUDA and Nvidia deployment experience does not substitute, because the entire job is the Intel toolchain and its quirks. If you have that background it is a rare seat, since few teams in India run this surface at production scale.
Apply on company site
Opens jobs.ashbyhq.com, the employer's own application page. Applying is always free.