Solutions Engineer / Sales Engineer

Overview

You're the technical expert on sales calls for Modular's AI infrastructure stack. When an AE gets a prospect interested, you take over the technical deep-dive: understanding their current ML infrastructure, running demos of MAX/Mojo/BentoML in their environment, building POCs that prove performance gains, and troubleshooting integration issues. Post-BentoML acquisition, you're showing how the full stack (optimization + serving) works together.

Role Snapshot

Company Context

Stage: Series B+ (315 employees, post-acquisition growth mode)

Size: 315 employees

Growth: Just acquired BentoML - SE team likely expanding to support larger product surface area and increased inbound from open source community.

Market Position: Category creator selling "hypervisor for AI compute." You're evangelizing new architecture patterns, not just feature-comparing against competitors.

GTM Reality

Pipeline Sources:

• Technical inbound from BentoML open source users hitting scale issues
• Outbound to companies with large ML infrastructure investments
• Community-driven leads from Mojo developers trying to go production

SE Team Structure: Likely small team (5-10 SEs) covering different verticals or regions. You might specialize in certain tech stacks (PyTorch vs. TensorFlow) or deployment patterns (cloud vs. on-prem).

Collaboration: You pair with AEs on deals. AE owns the commercial relationship; you own technical validation. You also work closely with Product/Engineering to feed back customer requirements and escalate bugs.

Competitive Landscape

What You're Up Against:

• Status quo (they built their own serving layer - "it works, why change?")
• Ray/Anyscale (distributed compute story)
• Cloud-native ML platforms (SageMaker, Vertex AI)
• DIY open source stacks (TorchServe, TensorFlow Serving, Triton)

Your Technical Edge:

• Hardware portability (same code on NVIDIA + AMD)
• Performance benchmarks (show quantified speedup)
• Enterprise hardening of BentoML open source

Common Technical Objections:

• "Our models are in [framework X], does this work?" (compatibility questions)
• "What's the performance overhead of your abstraction layer?" (they assume portability = slower)
• "We have 200+ models in production, how do we migrate?" (operational risk)
• "What if we hit a bug in production?" (maturity/support concerns)

What You'll Actually Do

Time Breakdown

POCs/Demos (50%) | Pre-sales Discovery (25%) | Internal (15%) | Learning (10%)

Key Activities

• Technical discovery calls: You hop on after the AE qualifies interest. You're asking: What models are you running? What's your current serving infrastructure? What frameworks (PyTorch, TensorFlow, JAX)? Where's it deployed (AWS, GCP, Azure, on-prem)? What's painful? (latency, cost, DevOps overhead). You're building a mental map of their architecture.

• Live demos in their environment: Not canned demos - you're actually deploying MAX/BentoML in their AWS account or on their hardware. You show their actual model running through your stack, benchmark latency/throughput vs. their current setup, and walk through how deployment would work in their CI/CD pipeline.

• Building and managing POCs: Once they commit to a proof of concept, you're hands-on for 30-60 days. This means: containerizing their models with BentoML, optimizing inference with MAX, deploying to their Kubernetes cluster, working with their DevOps team on integration, running load tests, and documenting results. You're troubleshooting errors, answering Slack questions, and adjusting scope when things don't work as expected.

• Creating technical collateral: You build reference architectures, deployment guides, and performance benchmark reports. These are for internal enablement and to send prospects. When a customer asks "how did [Company X] deploy this?," you point them to a case study you helped write.

The Honest Reality

What's Hard

• Every POC is custom: There's no cookie-cutter demo. Each company has different models, frameworks, deployment environments, and constraints. You're doing real engineering work under sales timelines. When something breaks during a POC, you're debugging in their prod-adjacent environment while the deal clock ticks.

• You're teaching a new category: Prospects don't understand "AI infrastructure hypervisor" yet. Half your job is educating senior engineers on why they should care about hardware portability or why Modular's approach is better than building in-house. Lots of "this is cool but we'll stick with what we have."

• Dependency on their team: POCs fail not because the tech doesn't work, but because their ML engineer got pulled onto another project, or security review took 6 weeks, or they deprioritized the evaluation. You're constantly nudging stakeholders to stay engaged.

• Rapid product evolution: Post-acquisition, the product surface area just doubled (BentoML integration). You're learning new components, updating demos, and answering "how does this work with [new feature]?" while also supporting active POCs.

What Success Looks Like

• Your POCs convert to deals at 50%+ rate (vs. industry average of 30-40%)
• Customers deploy to production within 90 days post-purchase (your POC de-risked the implementation)
• AEs request you specifically for their strategic deals (you're a trusted technical advisor)
• You contribute to 8-12 closed deals per year ($3M-5M in influenced ARR)

Who You're Selling To

Primary Buyers (Technical):

• ML Platform Engineers (hands-on with your product during POC)
• MLOps Leads / Staff Engineers (technical champions who push internally)
• Engineering Managers / Directors (decide whether to move forward with eval)

Secondary Stakeholders:

• DevOps/Infrastructure teams (integration concerns)
• Security teams (review your deployment model)
• CTO or VP Engineering (final approval on strategic infrastructure changes)

What They Care About:

• Does it actually work?: They've seen a lot of AI vendor hype. You need to prove performance gains with their real models, not synthetic benchmarks.
• Integration effort: How much work to migrate? Do they rewrite code? How does it fit their CI/CD?
• Operational risk: What happens if it breaks? Can their team debug it? What's your support model?
• Lock-in: They're trying to avoid vendor lock-in (currently NVIDIA). They want to make sure Modular doesn't become a new lock-in.
• Community/longevity: Is this a science project or production-grade? BentoML's 10K+ user base helps here.

Requirements

• 3-5+ years as Solutions Engineer, ML Engineer, or DevOps Engineer (preferably in ML infrastructure)
• Hands-on experience deploying ML models to production (PyTorch, TensorFlow, model serving frameworks)
• Strong Python skills - you're writing code during POCs, not just presenting slides
• Kubernetes/Docker experience (most customers deploy on K8s)
• Familiarity with cloud platforms (AWS, GCP, Azure) and GPU compute
• Comfortable with deep technical conversations with staff engineers and architects
• Bonus: experience with Ray, Triton, TorchServe, or other ML infrastructure tools
• Bonus: contributed to open source ML projects