Sign In Sign up free

Loading…

Terms of Service Privacy Policy Refund Policy

© Trevo.work

Applied Data Science at Backslash · Trevo

‹ Back to jobs

B

Applied Data Science

Verified Active

Tel Aviv-Yafo · Cyber · Senior

✓ Direct ATS✓ Not duplicate✓ Fresh

Match score

See how well you match this role, and what's worth sharpening before you apply.

Register to check your fit ›

Register to apply ›

Everything to know before you send your CV

We've gathered AI tools to check your fit, sharpen your CV, and prep for the interview.

Company careers page

We couldn't find a direct apply link for this specific role. Browse the company's official careers page for open positions.

Visit the company's careers page

Is this role relevant for you?

Full company profile

B

Stable hiringAttractivenessCStability~D

More roles at this company

Job description

About the role

Applied Data Scientist — Intent Drift Detection Backslash Security · Realtime AI-agent protection About Backslash We are builders who want to secure the next generation of software development from the inside out. Software development has entered a new era defined by vibe-coding, MCP servers, and AI agents. While these tools allow developers to build faster than ever, they also introduce a completely new frontier of security challenges that exist directly on the engineer's machine. Backslash is building the security layer for this new reality. We provide the essential guardrails and visibility needed to secure AI-driven workflows. By actively protecting developer computers and monitoring IDE-based AI interactions, we ensure the future of coding is as secure as it is fast.

What you'll do

AI coding agents run with enormous privilege on developer machines. When untrusted content steers one off its task, classical security sees only the final action — never that the agent's intent drifted. Detecting that drift, in realtime, on-device, is an open problem. No one has solved it. You will.
Build the Dataset: Design and generate the data that defines what "drift" looks like, and co-design the pipeline that creates it.
Train Small, Ship Local: Fine-tune, distill, and shrink the models that detect drift in realtime — small enough to run fully on-device, no cloud.
Shape the Architecture: Decide what's a model, what's a rule, and what's a gate. The detector's design is your call.
Measure What Matters: Build the evaluation harness — realtime detection at very low false-positive rates on a developer's machine is the product.
You own intent-drift analysis as a problem, not a ticket.

Requirements

Applied Data Science End-to-End: 3–5 years turning ambiguous problems into working models — problem framing, data collection and labeling strategy, feature/representation design, training, and evaluation. You've owned a dataset and the model that consumed it.
Small-Model Fluency: Comfortable fine-tuning small open models locally — DeBERTa-class encoders, ≤8B decoders, LoRA, quantization, single-GPU workflows.
Applied-Research Mindset: You approach an open research problem with passion and a practical, ship-it attitude.
Independence: You want an impactful, strategic problem in LLM security — not a ticket queue.
Startup DNA: Can-do, fast, a true partner. Non-negotiable.
Local-first by design. Python, PyTorch, HuggingFace, ≤8B models.

Nice to have

Security Background: Prompt injection, MITRE ATLAS, agent security — or the hunger to learn it fast.
Adversarial ML: Experience with LLM evaluation, red-teaming, or building adversarial datasets.
Endpoint Software: Familiarity with software that runs locally on developer machines with a low resource footprint.

Tech stack

PythonPyTorchLLM

Full company profile ›

Senior Full Stack EngineerTel Aviv-Yafo

Senior Software Engineer / Tech-LeadTel Aviv-Yafo

Get tech jobs on Telegram

Pick a channel for your field and get fresh jobs every day.

Telegram channels