You need someone who can ship a model to production, not just talk about one. Kiyansh Group places senior, US-based data science and ML talent on contract, contract-to-hire, and direct terms — screened by people who read the code, not just the resume.
Tell us what you need →Kiyansh places contract, contract-to-hire, and direct data science and ML professionals for teams that need Python, production ML pipelines, and applied LLM work. Every candidate we submit is US-based and works directly with your team. We do not run an offshore delivery center behind the scenes and pass off the work — the person you interview is the person who does the job, in your time zone, in your standups.
We staff across the range you actually hire for: data scientists who own the modeling and experimentation, ML engineers who build the training and serving pipelines, and LLM-focused engineers doing RAG, fine-tuning, and evaluation. Where a role fits an H-1B contractor, we place one; seniority and technical fit drive the submission, not visa category.
As a firm we also work as a sub-vendor to MSPs and prime vendors on private-sector and government engagements, so we're comfortable slotting into an existing vendor structure, badging process, or security requirement without friction.
Most staffing firms keyword-match a resume and forward it. We don't. A candidate for a data science or ML role goes through a technical screen run by an engineer who has done the work — a live conversation about a real problem, not a trivia quiz. We probe how they'd frame a modeling problem, handle leakage and imbalanced data, validate results, and reason about why a model regressed in production.
For engineering-heavy roles we review actual work: a code sample, a GitHub history, or a walkthrough of a pipeline they built — how they versioned data, managed features, tested, and monitored drift. For LLM work we dig into retrieval design, evaluation (how do you know the RAG answer is right?), latency and cost trade-offs, and where they chose fine-tuning over prompting and why.
We also screen for the parts that break contracts: can they communicate to non-technical stakeholders, do they scope honestly, and have they shipped something that real users touched. If a candidate can't defend their own decisions in a technical conversation, they don't get submitted to you.
The baseline is strong Python: NumPy, pandas, scikit-learn, and the ability to write code another engineer can maintain. On the modeling side we screen for real experience with PyTorch or TensorFlow, gradient-boosting libraries (XGBoost, LightGBM), and honest statistical judgment about evaluation and experiment design.
For production ML the signals that matter are pipeline and lifecycle tooling — MLflow or Weights & Biases for tracking, orchestration with Airflow or similar, containerization with Docker, and cloud fluency on AWS (SageMaker), GCP (Vertex AI), or Azure ML. For LLM roles we look for hands-on work with the OpenAI or Anthropic APIs, orchestration frameworks like LangChain or LlamaIndex, vector stores (pgvector, Pinecone, FAISS), and a working understanding of embeddings, retrieval, and evaluation.
Certifications like AWS Machine Learning Specialty or a cloud ML credential are a useful signal, not a substitute — we weight demonstrated, shippable work above any cert. If your stack has a hard requirement, tell us and we screen specifically against it rather than submitting near-matches.
We work three ways: contract for defined project or capacity needs, contract-to-hire when you want to evaluate a person before converting them, and direct placement for a permanent hire. Contract and contract-to-hire are the fastest to stand up because there's less internal approval overhead on your side.
Time-to-fill depends on the specificity of the role and your interview availability. For a well-defined contract data science or ML role, expect qualified, pre-screened candidates within roughly one to two weeks; niche LLM or domain-specific requirements take longer because we hold the bar rather than pad the pipeline. The constraint is usually candidate quality and your interview slots, not our sourcing.
We'd rather send you two people who fit than ten you have to filter. That means fewer submissions, each one already screened against your actual requirements — so your engineers spend interview time evaluating strong candidates instead of rejecting resume spam.
US-based, working directly with your team in your time zone. Kiyansh does not run a hidden offshore delivery model — the person you interview and hire is the person doing the work. Where appropriate we place H-1B contractors, but seniority and technical fit drive every submission.
An engineer who has done the work runs a live technical screen: framing a modeling problem, handling leakage and validation, reviewing real code or a pipeline they built, and — for LLM roles — probing retrieval design, evaluation, and cost trade-offs. If a candidate can't defend their own decisions, we don't submit them.
For a well-defined role, expect pre-screened candidates within roughly one to two weeks. Niche LLM or domain-specific requirements take longer because we hold the technical bar. Contract and contract-to-hire stand up fastest; the usual bottleneck is your interview availability, not our sourcing.
Tell Kiyansh Group what you're building and the stack it runs on, and we'll send screened, US-based data science and ML candidates who fit — not a stack of resumes to filter.
Start a conversation →