Machine Learning Engineer

• Create evaluation sets and automated evals for form-fill accuracy, match quality and agent success rate, then iterate on prompts, models and fine-tunes.
• Train, fine-tune and benchmark models (PyTorch), for example to parse CVs and job descriptions and to classify job fit.
• Deploy models and agents as scalable APIs and services on the cloud (AWS/GCP) with Docker, Kubernetes and CI/CD, integrated with our Node backend.
• Add observability: tracing of agent runs, latency/cost dashboards, failure and drift monitoring.
• Build data pipelines that collect and clean job listings from many sources.
• Work with product and with candidates to turn real job-seeker problems into shipped features, and care about responsible AI: candidates’ data, fairness and accuracy.

• Strong Python and solid software-engineering fundamentals (APIs, backend services, testing). (Python appears in 64% of the sampled postings; APIs/backend in 54%.)
• Hands-on experience building with LLMs: OpenAI/Anthropic APIs, prompting, agents or RAG. (LLMs 60%, agents 69%.)
• Experience shipping ML or AI systems to production and running them at scale. (Production 63%, scale 67%.)
• Experience training and evaluating models, plus familiarity with PyTorch. (Training/experimentation 48%.)
• Cloud experience (AWS, GCP or Azure) plus Docker/Kubernetes. (Cloud 40%.)
• A degree in Computer Science or a related field, or equivalent experience.
• A product mindset: you care about users and measurable impact. (36%.)

Desirable

• Building LLM evaluation frameworks (evals, benchmarks, LLM-as-judge). (40%)
• Monitoring and observability tooling (e.g. Datadog, OpenTelemetry, LangSmith). (55%)
• Open-source contributions or a public portfolio of AI projects. (36%)
• A second language such as TypeScript, Go, C++ or Rust. (TypeScript/JS 31%, Go 16%)
• GPU / CUDA / inference-performance optimisation. (22%)
• Fine-tuning (LoRA, RLHF/DPO). (18%)
• Familiarity with AI safety or responsible AI. (21%)
• Recommender/ranking systems, MLOps tooling (MLflow, model registry), web scraping/data pipelines. (13-15% each)