AI Integration Engineer
ey
Job Description
- Build end‑to‑end AI/ML pipelines (training → evaluation → deployment) using MLflow/Kubeflow/Databricks/Weights & Biases with experiment tracking and model registries.
- Develop models with Python using PyTorch, TensorFlow, JAX, scikit‑learn, and Hugging Face Transformers, package as reproducible services.
- Implement LLM/RAG systems with LangChain, LlamaIndex, Semantic Kernel and vector DBs (Pinecone, Weaviate, Milvus, FAISS, Chroma) for semantic retrieval and grounding.
- Fine‑tune and optimize models using PEFT/LoRA/QLoRA, DeepSpeed/Accelerate, distillation, and quantization; export/optimize via ONNX Runtime/TorchScript/TensorRT.
- Engineer scalable model serving with KServe, Seldon Core, BentoML, Ray Serve, NVIDIA Triton, supporting A/B, canary, shadow deployments.
- Build evaluation harnesses (offline/online) with Ragas, TruLens, Promptfoo, golden datasets, and regression gates integrated into CI/CD.
- Construct feature stores (e.g., Feast) and data contracts (Protobuf/Avro/Pydantic); enforce data quality with Great Expectations/Deequ.
- Orchestrate event‑driven pipelines with Airflow/Prefect/Dagster; streaming/messaging via Kafka/RabbitMQ/NATS and schema registries.
- Design Python microservices using FastAPI/gRPC; integrate with downstream systems via REST/GraphQL; write robust automation in Python/Bash/PowerShell and SQL for data ops.
- Use notebooks (Jupyter) and packaging (Poetry/pip/conda) with virtualenvs, environment locking, and artifacts suitable for promotion across stages.
- Apply testing & quality: pytest, unit/integration/e2e tests, property‑based (hypothesis), linters/formatters (ruff/flake8, black), type checks (mypy/pyright), pre‑commit.
- Deliver IaC with Terraform/Pulumi; manage config via Helm/Kustomize; implement GitOps with Argo CD/Flux on managed/self‑hosted Kubernetes.
- Build secure CI/CD (GitHub Actions/GitLab CI/Jenkins/Azure DevOps) for app/data/ML artifacts, artifact promotion, provenance, and automated rollbacks.
- Embed DevSecOps: SAST/DAST/IAST (Snyk/Checkmarx/SonarQube), container & IaC scanning (Trivy), dependency hygiene (Dependabot/Renovate), SBOM (Syft/CycloneDX).
- Enforce policy‑as‑code (OPA/Gatekeeper, Kyverno), image signing/verification (Sigstore/cosign), supply‑chain standards (SLSA, in‑toto).
- Manage secrets/KMS with Vault and native managers; adopt short‑lived workload identities, mTLS, and least‑privilege RBAC/ABAC in clusters and pipelines.
- Implement AI safety & governance: prompt‑injection defenses, output filtering, PII redaction, guardrails (Guardrails.ai/NeMo Guardrails/Presidio), policy checks.
- Monitor model/data drift, bias, and performance with Evidently/WhyLabs/Arize/Fiddler; unify telemetry via OpenTelemetry, Prometheus, Grafana, ELK/Loki, Jaeger.
- Optimize compute/GPU: CUDA/cuDNN/NCCL, HPA/VPA/KEDA, efficient batching, caching, concurrency control; track cost and latency SLOs.
- Implement progressive delivery for services/models (blue/green, canary, shadow) using Argo Rollouts/Flagger with instant rollback and health checks.
- Operate API gateways and service mesh (Kong/NGINX/Envoy, Istio/Linkerd) for rate limiting, mTLS, authN/Z, and zero‑trust patterns.
- Ensure privacy/compliance (GDPR/CCPA/DPDP/ISO 27001): data minimization, masking/tokenization, DLP, lineage (OpenLineage/Marquez), model cards/data sheets.
- Collaborate with security, data, and platform teams to publish golden paths, templates, and reference implementations for repeatable AI delivery.
- Contribute to code/design reviews and SRE practices (SLIs/SLOs/error budgets), on‑call readiness, incident response, and blameless post‑mortems.