משרה 10763 | דרוש עובד/ת בתחום פיתוח תוכנה בh...

דרוש עובד/ת בתחום פיתוח תוכנה בhybrid

2026-05-17 16:45:48

פרטי המשרה

Funded AI security startup hiring — 2 ML roles We're an Israeli startup (funded pre-seed) building a real-time defense platform for AI and LLM systems — detecting prompt injection, jailbreaking, data exfiltration, and more. The product is already built and running — a multi-tier detection pipeline with trained ML classifiers, an AI Judge, embedding-based detection, and adversarial defenses. We're looking for the people who will take it to the next level. We're not looking for yes-men. We need people who challenge assumptions, push back when something isn't right, and own their domain. If you see a better way to detect attacks — speak up, prove it, ship it. Our goal is clear: maximize detection accuracy while keeping false positives near zero, at production-grade latency. Every percentage point matters. Role 1 — Senior ML / AI Security Engineer What you'll do Train and improve transformer-based classifiers (BERT variants, ONNX) for attack detection Red-team the system — try to break it, then build the defense Research and implement advanced defense techniques (adversarial training, perturbation voting, ensemble classifiers) Work closely with the CTO on detection architecture and model strategy Challenge the current pipeline — if you find a better approach, we want to hear it What you need Hands-on experience with PyTorch, NLP, fine-tuning transformer models Understanding of adversarial ML / LLM security (jailbreaking, prompt injection, GCG-style attacks) Ability to read and implement academic papers quickly Adversarial mindset — you think like an attacker Strong opinions, loosely held — you bring ideas, not just execution Role 2 — ML Data / Research Engineer What you'll do Build and maintain training datasets (1M+ samples, 15+ sources) Data augmentation — generate attack and benign examples using LLMs Benchmarking and regression testing — measure model performance and track improvements across versions Research and integrate new datasets from HuggingFace and academic papers Identify and fix data gaps — find the cases where the model fails and fix them Take ownership of data quality — if the model is wrong, the data is wrong, and that's your call What you need Strong Python, experience working with large datasets Familiarity with the HuggingFace ecosystem (datasets, tokenizers, models) Statistical thinking — understanding of class balance, precision/recall, FPR trade-offs Ability to write end-to-end data pipeline scripts Bonus: experience with NLP / transformers / LLM-based data generation You care deeply about getting it right — mislabeled data keeps you up at night Both roles Full-time, Israel (hybrid — one day/week in office) Salary + equity Small team (3 people), direct impact on the product from day one We want leaders, not followers. You'll own your domain end-to-end. If you're excited about building the security layer for the AI era — and you want to actually shape the product, not just write code someone else designed — we want to talk..