← Back to All Openings

Ground-Truth Observability Layer Systems Architect

corpora-jobs-1778796293285-db9d41c6 - Frontier Development
Systems EngineerStaff1 position
⚡

Why This Role is Different

Frontier Development Role

Lead the design and deployment of the world’s first real‑time, model‑agnostic observability layer that turns a black‑box LLM into a transparent, auditable engine. You’ll build the hardware‑software bridge that captures every internal state change and publish it to a tamper‑evident ledger, enabling instant detection of deceptive reasoning.

The Frontier Element

This role pushes the boundary of AI safety by marrying low‑latency hardware instrumentation with blockchain‑based attestation—an area where no production system currently exists. Your work will set a new standard for internal state observability in large‑scale AI systems.

🔬

Project Context

Research Area

Ground-Truth Observability Layer (GLO) and Multi-Agent Verification Protocol (MAVP)

From: Adversarial Prompt Injection and Misleading Explanations

Why This Role is Critical

The GLO must capture every internal state change of a closed‑source LLM in real time, while MAVP requires a tamper‑evident ledger and hardware attestation. Both demand a unified, ultra‑low‑latency, hardware‑centric system that can operate outside the model’s inference loop.

What You Will Build

A dedicated sensor stack (GPU/TPU hooks, kernel‑level hooks, or custom ASIC) that streams attention maps, token embeddings, and logits to a distributed ledger; a Merkle‑tree blockchain with cryptographic signatures; and an end‑to‑end monitoring dashboard that flags state divergences within milliseconds.

🛠

Key Responsibilities

  • Architect and prototype a low‑latency sensor that hooks into GPU/TPU inference pipelines without disrupting throughput.
  • Design and implement a Merkle‑tree based blockchain that records signed explanation fragments and internal state snapshots.
  • Integrate hardware attestation (e.g., TPM, SGX) to guarantee that the sensor runs on trusted hardware.
  • Develop a real‑time monitoring dashboard that visualizes attention maps, logits, and ledger entries for operators.
  • Collaborate with LLM developers to expose internal hooks and validate sensor coverage across multiple model architectures.
🎯

Required Skills & Experience

Technical Must-Haves

Low‑latency systems programming (Rust/C++)

Expert
Implement kernel‑level hooks and real‑time data pipelines.

GPU/TPU internals and inference engine architecture

Advanced
Hook into transformer layers to capture attention and logits.

Blockchain and Merkle‑tree design

Expert
Build a tamper‑evident ledger for explanation fragments.

Hardware attestation (TPM, SGX, HSM)

Advanced
Guarantee sensor integrity on trusted hardware.

Distributed systems and real‑time data streaming

Expert
Ensure 99.9% coverage with sub‑10 ms latency.

Experience Requirements

  • 5+ years designing high‑performance, low‑latency systems for AI workloads.
  • Track record of building production‑grade instrumentation for large‑scale ML models.
  • Experience with hardware security modules and cryptographic protocols.

Education

PhD or Master’s in Computer Engineering, Systems, or a related field with a focus on real‑time systems or hardware security.

⭐

Preferred Skills

  • Experience with custom ASIC or FPGA design for AI inference.
  • Knowledge of privacy‑preserving ledger designs (e.g., zero‑knowledge proofs).
  • Prior work on LLM internals or transformer debugging.
🤝

You Will Thrive Here If...

  • Thrives in environments where you design, ship, and own a system end‑to‑end without hand‑offs.
  • Comfortable making rapid, data‑driven decisions in a high‑stakes, fast‑moving context.
📈

Impact & Growth

12-Month Impact

Within 12 months, deliver a fully operational GLO that captures >99% of internal state changes with <10 ms latency, and a MAVP ledger that can detect and flag deceptive explanation fragments in real time, reducing false‑negative jailbreaks by >90%.

Growth Opportunity

Lead the expansion of the observability stack to multimodal models, integrate with downstream safety modules, and shape the company’s AI‑trustability platform.

Ready to Push the Boundaries?

If this sounds like the challenge you have been looking for, we want to hear from you. We value what you can build over where you have been.