Peter Massih

I do ML research and build the software that ships it. Currently at NEC Labs America. Previously a software engineer (AI/ML) at Lombard Odier.

github · huggingface · linkedin · email

Peter Massih

Background

Currently
  • ML research @ NEC Labs America · Princeton · Summer 2026
Previously
  • Software engineer AI/ML @ Lombard Odier · Geneva · 2025–2026
  • ML research @ NEC Labs America · Princeton · 2025
  • Full-stack AI engineer @ Ageospatial (startup) · Lausanne · 2024
Education
  • MS Computer Science (Data Analytics & ML) @ EPFL · 2023–2026
  • Exchange @ ETH Zurich (Electrical Engineering + Computer Science) · 2022–2023
  • BS Communication Systems (Electrical Engineering + Computer Science) @ EPFL · 2020–2023

Publications

Three Modalities of Production AI in Financial Services: Retrieval, Generation, and Detection

Peter A. Massih · MSc Thesis, EPFL (VILAB) & Lombard Odier, 2026.

Reasoning with Pixel-level Precision: QVLM Architecture and SQuID Dataset for Quantitative Geospatial Analytics

Peter A. Massih, Eric Cosatto · arXiv:2601.13401, 2026. arXiv →

All publications →

Selected projects

3D-BlockGen: diffusion pipeline that generates buildable LEGO models from text prompts. Open-sourced code, model, and a 542k-sample dataset from Objaverse.

EPFL-TA-Meister: Llama-3-8B fine-tuned as an EPFL teaching assistant: SFT, DPO alignment, RAG over coursebooks, then 4-way quantization. Five models shipped to HuggingFace.

Wikispeedia Navigation Analysis: mixed-effects model over 76k+ gameplay paths, identifying semantic similarity as the dominant navigation strategy (p < 0.001).

Reinforcement Learning Mountain Car: DQN with random network distillation and a direction-aware throttle reward, ~33% faster convergence on Gymnasium's MountainCar-v0.

All projects →

Peter's Patches

Posts will appear here once I publish them.

All patches →