Hugging Face AI/ML Engineer / Open-Source Model & Agent Platform Engineer
Open a video or resource link, watch it, then check its box to track your progress and unlock what comes next.
- Build confidently on the open-model ecosystem and ship useful agent experiences
- Connect agents to tools and adapt open models for domain-quality behavior
- Operate high-performance open-model and agent systems with eval-driven rigor
Benchmark: Hugging Face Open-Source ML Engineer + ML Developer Tools + optimized-inference / agent-platform signals (current Sep 2026). Market refresh: Transformers/Datasets/Hub; Inference Providers + Endpoints; vLLM/SGLang/TGI/TEI; smolagents; MCP + tools; Gradio/Spaces; PEFT/LoRA + TRL; open-source contribution; agent evals; scalable serving, observability, security and cost/latency optimization.
Search Live Openings
We don’t have a direct job-board feed matched to this one yet, so here are real, live US job searches for “Hugging Face AI/ML Engineer / Open-Source Model & Agent Platform Engineer” instead.
Build confidently on the open-model ecosystem and ship useful agent experiences
Open-model application engineering + unified inference
Coverage: Hub / Transformers / Inference Providers / Python / OpenAI-compatible APIs / Model Selection / Embeddings
smolagents, code agents + tool-calling foundations
Coverage: smolagents / CodeAgent / ToolCallingAgent / Tool Design / Sandboxing / Hub + Spaces / Gradio
Connect agents to tools and adapt open models for domain-quality behavior
MCP, tools + interoperable agent ecosystems
Coverage: MCP Client/Server / ToolCollection / Structured Tool Outputs / Permissions / Agent Interoperability
Model adaptation + reproducible open-source workflows
Coverage: PEFT / LoRA / TRL / Fine-Tuning / Datasets / Revisions / Model Cards / Reproducibility / Community Review
Operate high-performance open-model and agent systems with eval-driven rigor
Production inference, endpoints + accelerator efficiency
Coverage: Inference Endpoints / vLLM / SGLang / TGI / TEI / Autoscaling / Private Endpoints / Throughput / GPU Efficiency
Agent evaluation, observability + trustworthy open systems
Coverage: Agent Evals / Task Success / Tool Correctness / Tracing / Regression / Safety / Model-Revisions / Cost-Latency-Quality
Role-Readiness Proof / Capstone
Build and operate an open-model agent product end to end. Select a model from the Hub and justify model/license/revision choices; expose it through Inference Providers or a dedicated Endpoint; build a smolagents CodeAgent or ToolCallingAgent with at least two governed tools including one MCP integration; add domain data plus a PEFT/LoRA or prompt/context adaptation experiment; publish a usable Gradio/Space interface; create a replayable eval suite for task success, tool correctness, safety and failure recovery; then load-test and tune the serving path for throughput, latency and cost, add logs/metrics, and document rollback, model-card and open-source contribution practices.
Proof Rubric
- Demonstrate reproducible model and dataset choices
- clean Python/library engineering
- correct tool and MCP contracts
- safe code/tool execution
- sound adaptation judgment
- high-quality open-source documentation and tests
- measurable agent task success
- eval/regression coverage
- diagnosable traces and endpoint telemetry
- defensible vLLM/SGLang/TGI serving tradeoffs
- autoscaling and cost/latency evidence
- security/license/supply-chain awareness
- and a clear record of how community feedback or production evidence changed the system