About H2LooP

H2LooP.ai is building next-generation sovereign coding models and context platforms for system engineers across embedded systems, robotics, UAV, avionics, automotive, and industrial IoT. Our core team comes from Google, Toshiba, Cisco, Bosch, Philips, and NXP. We aren't building generic chatbots; we are building engineering-grade foundation models that understand hardware constraints, datasheets, and real-time execution.

About the Role

We are looking for a Data Science / Research Intern who is obsessed with how LLMs reason over technical data. You will work directly with our ML Lead and Principal Research Scientists to transform messy, heterogeneous engineering artifacts (PDFs, C codebases, schematics) into high-quality training signals. This is a "data-centric AI" role where your work directly moves the needle on model performance.

⭐ What You’ll Work On

Data Synthesis & Curating: Build pipelines to extract and structure knowledge from complex technical documentation (PDFs, HTML, Markdown).
Synthetic Data Generation: Experiment with LLM-based techniques to generate high-quality instruction-tuning pairs for system engineering tasks.
Evaluation Frameworks: Help build domain-specific benchmarks (beyond HumanEval) to test if models understand timing, safety, and memory constraints.
Quality Filtering: Develop automated heuristics and ML-based classifiers to prune low-quality code or documentation from training sets.
Literature Review: Stay on top of the latest papers in LLM reasoning, code generation, and RAG to propose experiments for the team.

🎯 Required Skills

Strong Python skills and experience with data libraries (Pandas, NumPy, Scikit-learn).
Familiarity with LLM frameworks (Hugging Face, PyTorch, LangChain, or LlamaIndex).
Experience with NLP techniques (tokenization, embeddings, vector databases).
Ability to write clean, reproducible code for data processing pipelines.
A research-oriented mindset: you enjoy digging into why a model is failing.

🌟 Good to Have

Prior experience with Retrieval-Augmented Generation (RAG).
Knowledge of C/C++ or interest in how low-level software works.
Experience with fine-tuning open-source models (Llama, Mistral, Qwen).
Contributions to open-source ML projects or a strong Kaggle/GitHub profile.

🔧 Example Tasks

Scrape and clean an entire RTOS documentation set for model pre-training.
Build a script to convert 500+ pages of a SoC datasheet into a structured JSON for RAG evaluation.
Run an experiment to see if adding "Chain of Thought" prompting improves the model's ability to debug a C-pointer error.

Data Science / Research Intern – Domain Foundation Models