interview_prep_doc / art_AeltdAeQfHM

Content

{
  "markdown": "# Together AI Interview Prep\n## Questions to Ask Rochelle Mattern \u2014 Head of Field Engineering\n\n---\n\n## 1. Interviewer Summary\n\nRochelle Mattern is the Head of Field Engineering at Together AI, a role she stepped into in September 2025. Before Together AI, she led Worldwide Field Engineering and Solutions Engineering at SambaNova Systems \u2014 a company pushing hard on custom AI accelerators for inference and fine-tuning \u2014 giving her direct, hands-on exposure to the hardware-software co-optimization challenges that sit at the heart of this FDE role.\n\nHer career arc is unusually broad and grounded: she started in hardware and electrical engineering at Cisco and United Technologies, moved into Google Cloud customer engineering (where she earned President's Club recognition supporting large strategic accounts), and then transitioned into building and scaling technical pre- and post-sales organizations at AI-native companies. At Forethought, she achieved a 75% reduction in time-to-value through revamped onboarding and trial frameworks \u2014 a metric that maps directly onto the \"opinionated onboarding\" responsibility in this job description.\n\n**What this means for your conversation:**\n- She will likely probe your **customer-facing technical depth** \u2014 not just whether you know vLLM or GRPO, but whether you can translate that depth into customer outcomes\n- She cares deeply about **time-to-value** and structured onboarding; expect questions about how you get customers to production quickly\n- She has personally championed the **product feedback loop** \u2014 field insights that change roadmaps \u2014 so she'll want to see that instinct in you\n- She is herself relatively new to Together AI (roughly a year in), so frame questions as **forward-looking and collaborative**, not as audits of what she's already built\n\n**Shared context to lean into:**\n- Open-source LLM ecosystems and developer-facing platform thinking\n- Building 0-to-1 technical functions at AI-native companies\n- The tension between deep IC technical work and cross-functional influence across CX, Engineering, and Sales\n\nThis is likely to be a substantive, peer-level technical conversation \u2014 not a standard behavioral screen. Thoughtful questions will signal that you've done the work and are already thinking like an FDE.\n\n---\n\n## 2. Best Questions to Ask Rochelle Mattern\n\n---\n\n### A. Interviewer Experience and Rapport\n\n**Question 1**\n\n> **\"You built the field engineering function at SambaNova from Director to Head of Worldwide, and now you're doing something similar at Together AI. What was the biggest structural lesson from SambaNova that you're deliberately applying \u2014 or deliberately not repeating \u2014 as you build out this team?\"**\n\n**Why this is a good question:** This surfaces Rochelle's mental model for what \"good\" field engineering looks like at an AI infrastructure company. It signals you've studied her trajectory rather than just her current title, and it invites her to share the institutional thinking that will directly shape how this team operates \u2014 and how she'll evaluate you.\n\n---\n\n**Question 2**\n\n> **\"At Forethought you reduced time-to-value by 75% through revamped onboarding and trial frameworks \u2014 that's a striking number. How much of that playbook is transferable to Together AI's inference and post-training context, where the technical surface area is so much deeper?\"**\n\n**Why this is a good question:** This shows you read her profile carefully and invites her to articulate what \"opinionated onboarding\" means in practice at Together AI \u2014 which is explicitly listed as a core FDE responsibility. Her answer will tell you how structured or freeform the onboarding motion actually is, and where there's room to build.\n\n---\n\n**Question 3**\n\n> **\"You came up through Google Cloud customer engineering supporting some of their largest strategic accounts before moving into AI-native companies. How has your definition of a 'technical win' evolved as the products shifted from cloud infrastructure to GenAI platforms?\"**\n\n**Why this is a good question:** This builds genuine rapport around her Google tenure and gets her talking about how she evaluates FDE success \u2014 which is exactly the calibration data you need to understand what she'll measure you against in your first 90 days.\n\n---\n\n### B. Role and Team Dynamics\n\n**Question 4**\n\n> **\"The job description is explicit that FDE is not a replacement for a Solutions Architect \u2014 you're a deep-domain specialist partnering with SAs. In practice, where does that boundary get tested most, and how does the team navigate it when a customer engagement requires both breadths?\"**\n\n**Why this is a good question:** This is the most operationally ambiguous part of the role. Understanding the real division of labor prevents stepping on SA toes and clarifies where you'll actually spend your time. A vague answer here is itself useful signal.\n\n---\n\n**Question 5**\n\n> **\"What does success look like at 30, 60, and 90 days for this role \u2014 and is the expectation that I'm running independent customer engagements by day 90, or is there a ramp period where I'm shadowing existing accounts first?\"**\n\n**Why this is a good question:** This gives you practical signal on ramp expectations and whether there's a structured onboarding or a \"drink from the firehose\" culture. It also demonstrates that you're thinking about how to deliver value quickly \u2014 which maps directly onto her time-to-value orientation.\n\n---\n\n**Question 6**\n\n> **\"How many strategic accounts would a single FDE typically be aligned to at any given time, and what's the mechanism for deciding when an account graduates from needing FDE-level support to being handled by the SA or CSM layer?\"**\n\n**Why this is a good question:** Capacity and account management clarity are critical for understanding whether the role is sustainable and how impact is measured. This also surfaces whether there's a defined escalation/de-escalation framework or whether it's ad hoc.\n\n---\n\n**Question 7**\n\n> **\"The job description mentions FDEs contribute back to the product where needed. Can you give me a concrete example of a field insight that actually changed something on the inference or post-training roadmap in the last six months?\"**\n\n**Why this is a good question:** This tests whether the product feedback loop is real or aspirational. If she can cite a specific example, you know the channel exists and is respected. If she can't, that's equally important to know \u2014 and signals where you might need to build the muscle yourself.\n\n---\n\n### C. Technical Environment\n\n**Question 8**\n\n> **\"Together AI serves 400+ trillion tokens a month across a marketplace of open models. When an FDE is tuning KV cache or selecting tensor parallelism configs for a strategic customer, are they working directly in Together's inference infrastructure, or are they building configurations that get handed off to a platform engineering team to deploy?\"**\n\n**Why this is a good question:** This clarifies the actual technical access and ownership model. The difference between advising and doing is significant for someone with hands-on inference engine depth \u2014 and it directly affects how you'd position your vLLM/SGLang/TensorRT-LLM experience.\n\n---\n\n**Question 9**\n\n> **\"For post-training engagements \u2014 LoRA, DPO, GRPO pipelines \u2014 is the FDE running training jobs on Together's compute, helping customers run on their own infrastructure, or both? And how mature is the tooling around experiment tracking and reproducibility for customer fine-tuning runs today?\"**\n\n**Why this is a good question:** This is directly relevant to your RL workbench work \u2014 you've benchmarked GRPO, DPO, and PPO across TRL, VeRL, OpenRLHF, and NeMo RL and know exactly where the tooling gaps tend to appear. Her answer will tell you whether there's greenfield work to be done or a mature platform to operate within.\n\n---\n\n**Question 10**\n\n> **\"Which inference engines are most prevalent in your current strategic customer deployments \u2014 vLLM, TensorRT-LLM, SGLang \u2014 and is there an internal standard, or does the FDE genuinely need to be fluent across all three depending on the customer's hardware profile?\"**\n\n**Why this is a good question:** This is a practical depth question that signals you know the landscape and want to understand where to invest preparation time before day one. It also surfaces whether Together AI has strong opinions about engine selection or treats it as customer-driven.\n\n---\n\n### D. Culture and Working Style\n\n**Question 11**\n\n> **\"How does the field engineering team handle a situation where an FDE's recommendation to a customer conflicts with the current platform's capabilities \u2014 do you lean toward honest scoping of limitations, or is there pressure to find a workaround to protect the deal?\"**\n\n**Why this is a good question:** This tests for integrity and a customer-first culture versus sales pressure dynamics. It's a critical question for a role that explicitly sits at the intersection of CX, Engineering, and Sales \u2014 three organizations with occasionally competing incentives.\n\n---\n\n**Question 12**\n\n> **\"Together AI is clearly moving fast \u2014 400T tokens a month is significant operational scale. How does the team balance the urgency of supporting strategic POCs with the discipline needed to document learnings and avoid reinventing solutions across accounts?\"**\n\n**Why this is a good question:** This surfaces knowledge management culture and whether there's institutional memory or whether every FDE is starting from scratch on each engagement. Given your background building DevPortals and onboarding frameworks at Intuit, this is also a space where you could add real value \u2014 and her answer will tell you whether that's welcome.\n\n---\n\n**Question 13**\n\n> **\"When an FDE disagrees with a product or engineering decision that's affecting customer outcomes \u2014 say, a configuration limitation that's costing a strategic account performance \u2014 what's the actual escalation path, and how receptive is the engineering org to field-driven urgency?\"**\n\n**Why this is a good question:** This tests psychological safety and cross-functional influence \u2014 both critical for someone whose core value proposition is the product feedback loop. It also signals that you think of yourself as a systems-level contributor, not just an account-level fixer.\n\n---\n\n### E. Growth and Development\n\n**Question 14**\n\n> **\"Your own path went from individual contributor customer engineer at Google to building and leading global field engineering organizations. For someone joining as an FDE today, is there a defined path toward technical leadership or team-building, or is the expectation that this role stays deeply IC for the foreseeable future?\"**\n\n**Why this is a good question:** Anchoring this in her own career trajectory makes it feel like genuine curiosity rather than impatience. It gathers real signal on the growth ceiling while paying her a genuine compliment about the path she's built.\n\n---\n\n**Question 15**\n\n> **\"The inference and post-training space is evolving extremely fast \u2014 new architectures, new training paradigms, new engines every few months. How does Together AI invest in keeping FDEs at the frontier technically, beyond what they pick up from customer engagements?\"**\n\n**Why this is a good question:** This signals you're thinking about long-term technical currency, not just current competency. It also opens a natural door to discuss your own practice of building research workbenches and evaluation platforms to stay current.\n\n---\n\n### F. Strategy and Vision\n\n**Question 16**\n\n> **\"Together AI positions itself as the 'AI Native Cloud' with a marketplace of open models. As customers increasingly want to own and adapt models rather than just call APIs, how does the FDE role evolve \u2014 are you seeing the center of gravity shift more toward post-training and fine-tuning engagements relative to pure inference optimization?\"**\n\n**Why this is a good question:** This shows you've internalized the company's positioning and are thinking about where the field engineering function is heading \u2014 not just what the job description says today.\n\n---\n\n**Question 17**\n\n> **\"Together's customer list includes Cursor, Decagon, and ElevenLabs \u2014 these are AI-native builders, not traditional enterprises. Does the FDE motion look meaningfully different for AI-native customers versus more traditional enterprise accounts, and is the team investing in both segments equally?\"**\n\n**Why this is a good question:** This surfaces customer segmentation strategy and whether the role will require context-switching between very different buyer profiles. AI-native customers often move faster and have stronger opinions about tooling \u2014 knowing the mix matters for how you'd calibrate your approach.\n\n---\n\n**Question 18**\n\n> **\"With GRPO and reasoning model post-training becoming a major focus across the industry, how is Together AI positioning its platform for customers who want to replicate or build on top of reasoning-style training pipelines \u2014 and is that an area where FDEs are expected to be at the frontier?\"**\n\n**Why this is a good question:** This is directly relevant to your RL workbench work \u2014 you've implemented GRPO, DAPO, REINFORCE++, and RLOO and benchmarked them across frameworks. It signals deep awareness of the current post-training landscape and positions your hands-on experience as immediately applicable.\n\n---\n\n### G. Shared Context \u2014 Connecting Your Work to Their World\n\n**Question 19**\n\n> **\"I've been building a post-training RL workbench that benchmarks GRPO, DPO, and PPO across TRL, VeRL, OpenRLHF, and NeMo RL \u2014 one thing I kept running into was how much framework choice affects throughput and memory footprint on the same hardware. Is framework-agnostic benchmarking something Together AI's customers are actively asking for, or do most strategic accounts arrive with a strong framework preference already?\"**\n\n**Why this is a good question:** This builds direct rapport by connecting your hands-on RL workbench project to a real customer question. It demonstrates both technical depth and field-readiness \u2014 you're not just asking about the role, you're already thinking like an FDE.\n\n---\n\n**Question 20**\n\n> **\"My background includes a lot of developer platform work \u2014 SDKs, DevPortals, onboarding frameworks \u2014 and I've noticed that the hardest part of inference optimization for customers is often not the tuning itself but getting them to instrument and surface the right metrics to know what to tune. Is that a gap you're seeing in the field, and is there appetite for FDEs to contribute tooling or documentation that closes it systematically?\"**\n\n**Why this is a good question:** This connects your Intuit platform and SDK experience to a real field engineering pain point \u2014 positioning your PM and developer-tooling background as additive rather than tangential to a deeply technical role. It also signals that you think in terms of leverage and systematic solutions, not just one-off fixes.\n\n---\n\n## 3. Best Conversation Starters\n\nUse these in the first two minutes to establish rapport before moving into substantive questions.\n\n### On her hardware roots\n> *\"I noticed you started your career in electrical engineering at Cisco before moving into cloud and then AI \u2014 I'm curious whether that hardware foundation ever comes back around in conversations about inference accelerators and silicon-level optimization with customers.\"*\n\n### On opinionated onboarding\n> *\"Your Forethought stat of 75% reduction in time-to-value caught my eye \u2014 I've been thinking a lot about what 'opinionated onboarding' means for inference customers specifically, where the configuration surface area is so much larger than a SaaS product.\"*\n\n### On SambaNova\n> *\"I saw you were at SambaNova during a really interesting period \u2014 they were pushing hard on custom AI accelerators for inference and fine-tuning. I'd love to hear how that experience shaped your thinking about what strategic customers actually need from a field engineering partner versus what they think they need.\"*\n\n---\n\n## 4. Topics to Handle Carefully\n\n### Topic 1: Rochelle is relatively new to Together AI herself\n\n**The risk:** She joined in September 2025 \u2014 roughly a year in. Questions that assume deep institutional knowledge of Together AI's internal history, or that imply she should have already solved structural problems, can land awkwardly.\n\n**Ask this instead:**\n\n> *\"As you're building out the field engineering function here, what's the one thing you're most focused on getting right in the next six months?\"*\n\nThis is forward-looking and collaborative rather than implying a gap.\n\n**Softening phrasings to reuse:**\n- *\"As you're shaping this...\"*\n- *\"In the direction you're taking the team...\"*\n- *\"From what you've seen so far at Together AI...\"*\n\n---\n\n### Topic 2: Her tenure at SambaNova was relatively short\n\n**The risk:** Her time as Head of WW Field Engineering at SambaNova was brief before she moved to Together AI. Do not probe the reasons for the transition or draw comparisons between SambaNova and Together AI in a way that could feel like you're asking her to criticize a former employer.\n\n**Ask this instead:**\n\n> *\"What drew you to Together AI specifically \u2014 what about the platform or the team made this the right next move?\"*\n\nThis invites her to tell the story on her own terms.\n\n---\n\n### Topic 3: The three-org reporting structure (CX, Engineering, Sales)\n\n**The risk:** The FDE role spans Customer Experience, Engineering, and Sales \u2014 a three-way influence structure that can be politically complex. Questions that sound like you're trying to identify which org \"really\" owns FDEs can signal concern about org politics rather than genuine curiosity.\n\n**Ask this instead:**\n\n> *\"How does the field engineering team coordinate across CX, Engineering, and Sales to make sure customer insights actually land in the right places?\"*\n\nThis frames the question as curiosity about coordination and leverage, not about hierarchy.\n\n---\n\n## 5. Best Questions to Prioritize During the Call\n\nIf time is limited, these five questions will give you the highest-signal picture of the role, the team, and Rochelle's expectations.\n\n1. **Question 4 \u2014 The SA/FDE boundary.** This is the most operationally ambiguous part of the role; her answer will define what your day-to-day actually looks like.\n\n2. **Question 7 \u2014 The product feedback loop.** Tests whether the most distinctive part of the FDE value proposition is real or aspirational at Together AI.\n\n3. **Question 8 \u2014 Technical access and ownership.** The difference between advising and doing is the difference between two very different jobs; you need to know which this is.\n\n4. **Question 19 \u2014 Framework-agnostic benchmarking (shared context).** Connects your RL workbench directly to a customer-facing question and demonstrates you're already thinking like an FDE.\n\n5. **Question 5 \u2014 30/60/90-day success.** Practical ramp clarity and a strong signal that you're focused on delivering value quickly \u2014 which maps directly onto her time-to-value orientation.\n\n---\n\n## 6. Suggested Call Flow\n\nA natural sequence for a 45\u201360 minute conversation:\n\n### Opening \u2014 establish rapport (first 3\u20135 minutes)\nUse one of the conversation starters. The SambaNova opener tends to land well because it's specific, shows research, and invites a story rather than a yes/no.\n\n> *\"I saw you were at SambaNova during a really interesting period \u2014 I'd love to hear how that experience shaped your thinking about what strategic customers actually need from a field engineering partner.\"*\n\n### After she's shared her background \u2014 pivot to the role (minutes 5\u201315)\nMove into the structural questions about the role before getting technical. This establishes shared vocabulary.\n\n> *\"That context is really helpful. I'd love to understand how the FDE role fits into the broader field engineering motion here \u2014 specifically where the SA/FDE boundary tends to get tested in practice.\"*\n\n### When discussing the role \u2014 go technical (minutes 15\u201330)\nThis is where Questions 8, 9, and 10 belong. You've earned the right to go deep by now.\n\n> *\"On the inference side \u2014 when an FDE is tuning KV cache or selecting tensor parallelism configs, are they working directly in Together's infrastructure, or building configs that get handed off to platform engineering?\"*\n\n### Near the end \u2014 shared context and product vision (minutes 30\u201340)\nThis is where Questions 19 and 20 land best. They feel like a natural \"here's what I've been thinking about\" rather than a prepared question.\n\n> *\"One thing I've been running into in my own RL workbench work is how much framework choice affects throughput and memory footprint on the same hardware \u2014 is framework-agnostic benchmarking something your strategic customers are actively asking for?\"*\n\n### Closing \u2014 growth and a strong finish (last 5 minutes)\nEnd with Question 14 (career path) or Question 15 (staying at the frontier), then close with a strong line.\n\n> *\"Last question \u2014 the inference and post-training space is moving so fast. How does Together AI invest in keeping FDEs at the frontier technically, beyond what they pick up from customer engagements?\"*\n\n---\n\n## 7. Final Recommended Question Set\n\nIf you only have time for four to six questions, ask these:\n\n1. **Where does the SA/FDE boundary get tested most, and how does the team navigate it?** *(Role clarity)*\n2. **What does success look like at 30, 60, and 90 days?** *(Ramp expectations)*\n3. **When an FDE is tuning KV cache or tensor parallelism, are they working directly in Together's infrastructure or handing off configs?** *(Technical ownership)*\n4. **Can you give me a concrete example of a field insight that actually changed something on the inference or post-training roadmap?** *(Product feedback loop reality check)*\n5. **I've been benchmarking GRPO/DPO across TRL, VeRL, OpenRLHF, and NeMo RL \u2014 is framework-agnostic benchmarking something your strategic customers are asking for?** *(Shared context + technical credibility)*\n6. **What drew you to Together AI specifically?** *(Rapport + understanding her conviction in the platform)*\n\n---\n\n## 8. Strong Closing Line\n\nUse this \u2014 or a close variation \u2014 to end the conversation:\n\n> *\"I've spent the last year building post-training workbenches, evaluation platforms, and developer tooling from scratch \u2014 and what I keep coming back to is that the highest-leverage work happens at the intersection of deep technical depth and systematic customer enablement. That's exactly what this FDE role seems designed to do, and it's why I'm genuinely excited about this conversation. I'd love to find a way to contribute here.\"*\n\n---\n\n## 9. Final Reminder\n\nThis conversation should demonstrate that you are:\n\n- **Already thinking like an FDE** \u2014 your questions should feel like the questions a great FDE would ask on day one, not the questions of someone trying to learn what the job is\n- **Technically credible without being performative** \u2014 reference your RL workbench, your inference benchmarking, and your platform work naturally, in the context of questions, not as a r\u00e9sum\u00e9 recitation\n- **Customer-outcome oriented** \u2014 every technical depth signal should connect back to what it means for a customer's time-to-value or production performance\n- **A product feedback loop thinker** \u2014 signal that you see field insights as a lever for improving the platform, not just closing individual accounts\n- **A peer, not a supplicant** \u2014 Rochelle built her career by asking hard questions and building things from scratch; she will respect someone who does the same in this conversation",
  "source_artifact_id": "art_dT4yc1q9GdU",
  "source_kind": "candidate_questions",
  "doc_type": "questions_to_ask"
}