Member of Technical Staff - Multimodal

Focus on text-image-tool reasoning quality and multimodal failure diagnosis.

Role Context

This role focuses on multimodal reliability in real user workflows. You will diagnose where text, visual, and tool outputs diverge from ground truth, then convert those findings into deployable evaluation and mitigation plans.

• Own multimodal failure taxonomy and benchmark evolution.
• Design targeted tests for image reasoning, OCR-heavy tasks, and tool-use chains.
• Partner with model and product teams to reduce cross-modal error propagation.

First 90 Days

• Baseline multimodal error rates across priority user journeys.
• Deliver a failure analysis report with top mitigation opportunities.
• Ship an evaluation harness for at least one new multimodal risk category.

Cross-Functional Partners

• Vision and multimodal model engineering teams.
• Tooling teams for retrieval, OCR, and external API integration.
• Product and design teams for end-user interaction quality.

Success Metrics

• Decrease in multimodal failure frequency on targeted tasks.
• Coverage depth of multimodal benchmark suites.
• Adoption rate of shipped mitigations by product teams.

Responsibilities

• Design evaluations and diagnostics for multimodal model outputs.
• Investigate failure modes across text, image, and tool-augmented workflows.
• Prototype mitigations and translate findings into deployable improvements.
• Collaborate with engineering and product teams in fast iteration cycles.

Qualifications

• MS/PhD or equivalent experience in ML, computer vision, NLP, or computer science.
• Experience building and evaluating multimodal systems in production-like environments.

Compensation

Total compensation: USD $190,000 - $430,000 (base + equity + bonus, location dependent).