Skip to content
JobreachGet started

How to cold email for a data science or AI/ML job

AI hiring posts range from research-heavy data science to engineering roles that ship LLM features. Work out which one the post wants, then lead with one project that shows the same thing: a problem, a baseline, a measured result and, for engineering roles, something running.

Updated

What AI and data science posts ask for

Read the post for which of three jobs it really is; the titles overlap:

  • Data scientist: statistics, experiments, models with scikit-learn or XGBoost, SQL, and explaining results to the business.
  • ML engineer: training and serving models with PyTorch, data pipelines, deployment with FastAPI and Docker, and monitoring models in production.
  • GenAI or AI engineer: building on LLM APIs, retrieval (RAG) with a vector database, evaluating prompts and outputs, agents and tool use, often with LangChain or LlamaIndex.
  • Across all three: Python, Git, a cloud and, for some roles, a master’s degree, papers or a Kaggle record.
  • For interns, many AI startups ask for “a project you built end to end” rather than a list of courses.

Your proof as a fresher

Lead with one project that goes past the notebook:

  • An end-to-end project: real data, a simple baseline, your model, the metric that matters for that problem, and a deployed demo or API.
  • For GenAI roles: a RAG or agent system with a small test set you wrote, and how its answers changed as you changed retrieval or prompts.
  • A Kaggle competition placing: your rank out of how many teams, and what your solution did differently.
  • A paper, a thesis or a research internship, if the post asks for research.
  • An internship where a model of yours was used, even internally.

Name the metric and why it fits: recall for fraud, latency for a chatbot, forecast error for demand. A high accuracy with no baseline, on data where one class is rare, tells an ML lead you didn’t check.

Your proof if you are switching in

Most people move into ML from software engineering or data analysis. Lead with the half you already have:

  • From software engineering: the production side many ML candidates lack. APIs you built, pipelines you ran, latency you cut. Then one model you trained and deployed yourself.
  • From data analysis: the statistics side. Experiments you designed, forecasts you made and how far off they were. Then one model you trained and evaluated.
  • From research or academia: your paper or thesis in one plain sentence, and code someone else can run.

For GenAI roles, a backend engineer who has shipped one LLM feature to real users has exactly the proof many of these posts ask for. Lead with it.

Your proof in two sentences

Sentence one: the problem, the data and its size. Sentence two: your approach, the metric against a baseline, and where it runs.

Data science, fresher

In a Kaggle competition on forecasting store sales, I placed 212th of about 3,000 teams with a LightGBM model on lag and holiday features, after building a seasonal baseline first. My public notebook explains which features helped, which I dropped, and why.

ML engineering, switching from backend

As a backend engineer I built the API that serves our fraud model to 2 lakh transactions a day, with responses under 50 ms. In my own time I trained a fraud model on a public dataset, compared it with a logistic regression baseline at the same false-alarm rate, and deployed it the same way.

Both are made-up examples. The numbers in yours must be your own, and ones you can explain in an interview.

An example email

Kiran, a co-founder of an AI startup, posts for a GenAI engineer intern who has “built something with LLMs end to end”. Aditya is a final-year student. This is the whole email.

To
kiran@lexiloop.example
Subject
GenAI Engineer Intern: Aditya Verma, B.Tech 2026, RAG project

Hi Kiran,

I saw your post asking for a GenAI intern who has built with LLMs end to end. I’m a final-year B.Tech Computer Science student, and I have spent this year building and testing a retrieval system.

I built a question-answering assistant over 1,200 pages of my university’s rules and circulars, with pgvector for retrieval and a 120-question test set I wrote by hand. Re-ranking and smaller chunks took correct answers from 71 to 94 of the 120. It runs as a FastAPI service, and our exam office now uses it.

Lexiloop builds contract review on the same kind of retrieval, so your problems are the ones I want to go deeper on. My resume is attached, and I can start in January for six months.

Aditya Verma
+91 98xxx xxxxx
github.com/adityaverma-example

A made-up example: Aditya, Kiran and Lexiloop are fictional. Every claim in the email comes from Aditya’s own resume.

What makes it work:

  • It answers the post’s own words, “end to end”, with a project that has data, evaluation and users.
  • The test set and the before-and-after count show he measures, which is what separates an engineer from a demo.
  • The line about Lexiloop ties his retrieval work to theirs, from what their own site says.
  • His availability is plain: when, and for how long.

The subject line

The role, your name, your batch or years, and a two-word hint of your proof when the post asks for something specific.

  • GenAI Engineer Intern: Aditya Verma, B.Tech 2026, RAG project
  • Data Scientist: Ritu Sharma, 3 years in demand forecasting
  • ML Engineer: Varun Pillai, backend engineer, models in production

Avoid subjects like “Job application” or “Resume”. The person who posted may have several openings, and the role first tells them which one.

Mistakes to avoid

  • A list of every model family: CNN, RNN, LSTM, Transformers, XGBoost, SVM. Show one you used well.
  • Accuracy with no baseline and no word about how the classes were balanced.
  • A notebook link with no README, no results table and no way to run it.
  • Calling a thin wrapper around a chat API an “AI platform” on your resume. Say plainly what you built.
  • Sending an ML engineering post your best analysis project, or a data science post your deployment work. Match the job the post describes.
  • Listing a paper you have only read. Expect to be asked how it works.

And the ones that hurt in every role: a guessed address, a generic resume, and an email longer than a phone screen. More in how to email a recruiter about a LinkedIn hiring post.

Where AI and data science posts appear

  • LinkedIn posts by founders and engineers at AI startups, who often post openings themselves and ask for a GitHub link.
  • Company job boards on Greenhouse, Lever and Ashby, where many AI companies list their roles.
  • The Hacker News “Who is hiring” thread, which carries many remote AI and ML roles.
  • Research labs’ and universities’ own pages, for research assistant and research intern roles.

A LinkedIn post search to start from (search Posts, sorted by Latest):

AI and ML internships

("machine learning" OR GenAI OR "AI engineer" OR "data scientist") ("send your resume" OR "share your resume") intern

4 of 5 ORs and ANDs

Looking for a full-time role? Write fresher instead of intern, or drop the level word if you have experience. LinkedIn finds nothing at all once a search has more than five ORs and ANDs: see how to find hiring posts on LinkedIn. If there is no reply in about a week, follow up once.

Guides for other roles

All of them, with what to lead with for each: cold email by role.