AI/ML Engineer / Decision Scientist Intern

Summer Intern — AI/ML Engineer / Decision Scientist

About iCustomer

iCustomer is the Agentic Decision OS for Growth Marketing, fast growing, used by 100+ clients to turn audience intelligence into decisions and orchestration with learning loops built in. We're a venture-backed, US-headquartered startup, building the next category in growth marketing, the decision layer and helping growth teams optimize ROAS, CAC, and LTV. Customers don't just buy software from us, they buy outcomes.

The Opportunity

We're opening a summer internship for a standout AI/ML Engineer or Decision Scientist to work on one of two frontier projects your pick, based on fit and interest:

Track 1 — Causal AI for Marketing Decisions. Build the causal engine that powers iCustomer's Decisioning Waterfall. Think: causal graphs over marketing actions, do-calculus on real customer data, uplift modeling, and counterfactual reasoning baked into agentic workflows. The goal: move marketing from correlation to cause.

Track 2 — Domain-Specific SLMs for Marketing Science. Train and fine-tune Small Language Models specialized for marketing science tasks audience reasoning, attribution narration, campaign critique, FIRE scoring explanations. Distillation, RAG, evaluation harnesses, the whole stack. Ship a model that beats GPT-4-class general models on a narrow, high-value marketing benchmark.

Either track ships into the product and gets credit in published work — Substack, conference talks, and (if the work is strong) a co-authored paper or open-source release.

This is a real research + engineering role, not a check box internship project. You'll work directly with Iqbal Kaur (Head of Decision Science) and Abhi Yadav (Founder/CEO, MIT, multi-exit founder, early CDP category pioneer).

What You'll Do

• Scope the problem with Iqbal and the team. Design the experiment, build the dataset, ship the model.
• Live in notebooks (Jupyter / Colab / Claude Code). Run experiments, track them, reproduce them.
• Build evals — your model is only as good as the benchmark you measure it on.
• Integrate the output into iCustomer's iWorker stack so real customers feel the lift.
• Present your work weekly to the team. Write it up at the end of the summer.

 Who You Are

• Currently enrolled in an undergrad, Master's, or PhD program in Computer Science, Statistics, Applied Math, Operations Research, Decision Science, Economics (with ML), or equivalent or a recent grad (within the last 12 months).
• Strong in Python and the modern ML stack — PyTorch, Hugging Face, scikit-learn, pandas, numpy.
• Comfortable with research papers — you can read, replicate, and critique a recent NeurIPS / ICML / KDD paper.
• AI-native — Claude, ChatGPT, Cursor, Gemini are part of your daily workflow.
• A builder. You'd rather ship a working prototype in 2 weeks than write a 40-page proposal.
• Self motivated, work with minimum supervision and in a product environment

For Track 1 (Causal AI):

• Coursework or projects in causal inference, econometrics, or experimental design — DAGs, do-calculus, instrumental variables, propensity scoring, uplift modeling.
• Hands-on with DoWhy, EconML, CausalImpact, CausalNex, or PyMC.

For Track 2 (SLMs):

• Hands-on with fine-tuning, LoRA/QLoRA, distillation, or RAG on small open models (Llama, Mistral, Phi, Qwen, Gemma).
• Built and run eval harnesses (lm-eval-harness, custom benchmarks).
• A point of view on when small beats large.

Bonus Points

• A GitHub with real projects (not just course assignments).
• Published a paper, blog post, or open-source repo.
• Background in marketing science, ad-tech, recsys, or causal ML for decisions.
• Experience with agentic frameworks (LangGraph, DSPy, AutoGen, Claude Agent SDK).
• Won a Kaggle, MIT/Stanford competition, or hackathon.

Logistics

• Duration: 10–12 weeks, summer 2026.
• Location: US-based remote 

Mentors: Founders & CXOs with Decision Science expertise