What's the difference between senior DS and senior MLE in 2026?

Title shape varies by company; the day-to-day diverges. Senior DS at analytics-shape companies (Meta, Airbnb, Stripe, Netflix Studio, Uber Marketplace) is product-analytics-leaning: experiment design, causal inference, metric ownership, cross-functional partnership with PMs. Senior MLE at production-ML-shape companies (Google, Meta Ads, Netflix Recs, Apple Intelligence) is engineering-leaning: model training infra, deployment surfaces, monitoring, on-call. The two career paths converge again at staff+ where 'broad ML domain ownership' is the title regardless of analytics-vs-MLE history.

How important is on-call rotation at senior?

Significant at production-ML-shape companies. Senior MLEs at Netflix, Google, Meta Ads, and most large-scale production-ML teams are in the on-call rotation for their model-serving systems. The bar: ability to debug a p99-latency regression at 3am, identify whether it's a model issue or a serving-infrastructure issue, run the rollback playbook, write the post-incident document. Analytics-DS senior roles typically have a softer on-call shape — pager for data-pipeline failures or experiment-readout escalations, not 24/7 latency monitoring.

Should I move to an AI lab from FAANG at senior?

Depends on what you're optimizing for. AI labs (Anthropic, OpenAI, Scale AI, xAI, Cohere, Together) pay materially higher equity-heavy total comp at senior — commonly $700k–$1.4M+. The work is closer to research and frontier-model engineering. The risk: equity concentration in a single private company; less diversified than FAANG public stock. The career signal: AI-lab senior MTS is highly transferable to other AI labs and to staff-track roles at FAANG; a research-eng background opens doors that pure-application-ML doesn't. Worth the move for engineers who want frontier work and accept the equity-concentration risk.

How do I know I'm ready for staff promotion?

Three artifacts: (1) you've led one cross-team initiative whose outcome named you as load-bearing; (2) you've mentored at least one junior or mid engineer through a promotion under your sponsorship; (3) you have a documented strategic articulation — a one-pager or roadmap that ties your team's ML work to company-strategic outcomes. If all three are not in place 12 months before you go up for staff, you're not ready. Promotion takes 2–4 years from senior at most large tech companies.

What's the right approach to LLM / foundation-model work at senior?

Be conversational about the trade-off space, not the latest paper. The bar at senior in 2026: comfort fine-tuning open models at scale (FSDP / DeepSpeed Stage-3), opinion on prompt engineering vs PEFT vs full fine-tune for a given problem, fluency with the eval-harness ecosystem (lm-evaluation-harness, EleutherAI's framework at github.com/EleutherAI/lm-evaluation-harness), and an articulated view on RAG architectures (vector search vs dense retrieval vs graph retrieval). Reading every new arxiv paper is junior; having an opinion on which arxiv papers matter for your domain is senior.

Do I need to publish papers at senior?

Required at AI-lab research-track senior MTS. Optional at FAANG production-ML senior. Anthropic and OpenAI senior MTS roles often have a research-publication expectation; the public alignment-research and safety-research papers are co-authored by senior MTS. FAANG production-ML senior roles weight internal-impact (lift on the north-star metric) more than publications. Publishing one or two papers as a senior is a strong signal for staff-track promotion at FAANG and is increasingly expected at AI-labs.

Data Scientist / ML Engineer Hub

Senior Data Scientist / ML Engineer Guide for Tech Companies (2026)

By Blake Crosley · Last verified 2026-04-29

In short

Senior data scientist or ML engineer (5–8 years) at a tech company in 2026 owns a domain end-to-end: model architecture, eval methodology, productionization, monitoring, and stakeholder alignment for one substantive product area. FAANG-tier total comp clusters $380k–$580k at L5/E5/IC5; AI-labs (Anthropic Senior MTS, OpenAI Senior MTS) sit $700k–$1.4M+ on heavy equity. Senior is the level where you stop being graded on individual delivery and start being graded on the multiplier you create — juniors who level up under your mentorship, projects that ship because you unblocked the right people, and ML-system designs that other teams adopt.

Key takeaways

FAANG-tier senior DS / MLE total comp $380k–$580k at L5/E5/IC5 per levels.fyi 2026; Meta E5 DS $400k–$580k (levels.fyi/companies/facebook), Google L5 MLE $420k–$600k (levels.fyi/companies/google), Anthropic Senior MTS $700k–$1.4M+ (anthropic.com/careers + levels.fyi reports).
Senior owns ML system design at the team-or-org level: feature stores, training infra, eval harnesses, deployment surfaces. Building 'Designing Machine Learning Systems' (Chip Huyen, O'Reilly 2022) competency is mid-level prep; senior is operating from it daily.
LLM / foundation-model production fluency is non-negotiable: PEFT fine-tuning at scale (FSDP / DeepSpeed Stage-3), eval-harness design (lm-evaluation-harness, github.com/EleutherAI/lm-evaluation-harness), inference deployment (vLLM, TensorRT-LLM, Anthropic / OpenAI APIs).
Mentorship is the primary multiplier signal at senior. Promotion to staff (typically 2–4 years from senior) is bottlenecked on visible mentorship outcomes — juniors who level up directly under your sponsorship, with named contributions you scoped and unblocked.
Causal inference at senior+ is increasingly the differentiator from 'just ML' work — propensity scoring, IV, DiD, synthetic control. Susan Athey's NBER methods (athey.people.stanford.edu) and Hernán & Robins (hsph.harvard.edu/miguel-hernan/causal-inference-book) are the canonical references at FAANG analytics-DS senior+ roles.

What senior DS / MLEs actually do

Senior is the level where the work transitions from 'shipping individual projects' to 'compounding through other people.' The senior signal at most large tech companies in 2026 is composed of four behaviors:

Multi-project ownership. You own a domain — recommendations, search relevance, content moderation, fraud detection, or a foundation-model capability — not a single project. Multiple juniors and mid-level engineers report into your work; you scope their projects, review their analyses or experiments, and unblock them. Quantitatively: you have 3–5 projects in flight at any given time, with you as the technical lead but not the implementer on most.
Cross-team coordination. The senior signal is when other teams come to you with ML-shaped questions because your domain depth is recognized. You're brought into design reviews on adjacent teams. You write technical documents that get circulated org-wide. Your name shows up on architecture-review meetings outside your immediate team.
Mentorship that scales. Mentorship is the primary multiplier. Senior+ engineers who don't mentor stall at senior. The signal: a junior on your team levels up to mid in 18 months under your sponsorship; their promotion case names your scoping and review as load-bearing.
Strategic articulation. You can write a one-pager that ties the team's ML roadmap to a company-strategic outcome, not just to engineering metrics. PMs and engineering directors quote your one-pagers. This is the single largest gap between 'senior' and 'staff' at most companies — staff engineers do this for the org, senior engineers do it for their domain.

What senior IS NOT: 'senior implementer.' A senior who has the multiplier of one (themselves) is at risk of being managed out, not promoted. Staff promotion specifically demands evidence that the engineer's scope is a multiplier on others — projects that shipped because of your unblocking, juniors who leveled up under your mentorship, ML systems that other teams adopt.

ML system design at senior: a worked Netflix-shape example

A worked example — a senior MLE at a streaming company driving a foundation-model upgrade for the recommendation surface over a 6-month project:

Months 1–2: Scope and architectural decision. Existing system: a two-tower DLRM trained on user-item interactions, refreshed daily, serving 200M+ users at p99 ~80ms. The engagement metric has plateaued; the team has tried minor architectural changes without lift. Senior engineer's scope: investigate whether a foundation-model embedding (text descriptions of titles + tags + user history summaries) can produce a quality lift. Three approaches sketched: (a) replace the user tower with a transformer encoder, (b) add a foundation-model embedding as a feature into the existing two-tower, (c) build a separate retrieval lane that complements the two-tower.
Months 2–3: Offline experimentation. Build the eval-set: 10k held-out user-sessions with ground-truth completion. Approach (b) wins on simulated lift +6% relative to baseline; approach (a) is +4% but at 4x the inference cost; approach (c) is +9% but adds a serving lane. Decision: ship (b) as the v1, set (c) up as the v2 roadmap. Document the decision with the offline numbers, the cost projections, and the eval-design rationale.
Months 3–4: Production migration. Build the embedding pipeline (a fine-tuned Qwen-2.5-7B with LoRA on internal title metadata, embedding dimension 768, 200ms p99 batch latency at the catalog scale). Add the embedding as a feature in the two-tower. Re-train the two-tower on 30 days of historical interactions. Senior engineer scopes 4 sub-projects to mid and junior teammates: (1) the embedding pipeline (mid), (2) the feature integration (junior), (3) the eval harness (mid), (4) the monitoring layer (junior). Senior reviews each design doc, leaves dense feedback, unblocks technical decisions.
Months 4–5: A/B test. 1% rollout for 14 days, then 10% for 28 days. Primary metric: 28-day retention. Secondary: completion rate, content-discovery breadth. Result: +1.4% on 28-day retention (95% CI [+0.9%, +1.9%]); +3.2% on completion. Senior engineer writes the readout, defends the conclusion in metrics review, and writes the migration-and-rollback runbook.
Month 6: Rollout and write-up. 100% rollout. Internal tech-talk on the design + production migration. The junior who built the monitoring layer presents the drift-detection methodology at the company's monthly ML-org talk. The mid engineer who built the embedding pipeline becomes the named-owner of the embedding service. Senior engineer writes the staff-promotion-track artifact: a one-pager that names the engagement lift, the cost (~$8k/month additional inference), and the team-multiplier (4 sub-projects shipped under their scoping).

What made this senior scope: the engineer scoped, designed, and delivered a model-architecture upgrade end-to-end touching offline experimentation / production migration / A/B testing / monitoring with 4 sub-engineers reporting into the work. The same problem at mid level would have been one engineer shipping the integration, with senior overseeing. The same problem at staff level would have been the engineer scoping a 12-month foundation-model strategy across multiple recommendation surfaces, with senior engineers like this one as the leads.

The interview at senior: what gets tested

Senior interview rounds at FAANG-tier and AI-labs in 2026 share a common shape: 1 phone screen + 5–6 onsite rounds (1 ML coding, 1–2 ML system design, 1 stats / eval methodology, 1 behavioral / cross-functional, 1 team-match / hiring-committee at some companies). Senior-specific weighting:

Company	ML system design	Behavioral / leadership	Coding bar
Meta DS (E5)	2 rounds — product analytics + experimentation	Heavy — partnership with PM and eng leads	SQL deep + Python pandas
Google MLE (L5)	1–2 rounds — ML system + distributed-systems-leaning	Moderate — Googleyness	Highest algorithmic bar
Netflix MLE (L5)	2 rounds — production-ML scale + recommendation depth	Heavy — Netflix culture deck	Strong
Anthropic Senior MTS	2 rounds — research-engineer + eval-design	Heavy — research collaboration	Moderate; less leetcode
OpenAI Senior MTS	2 rounds — ML system + research-engineer	Heavy — research collaboration	Moderate
Databricks MLE (L5)	2 rounds — ML system + distributed ML	Moderate	Strong

Senior-specific questions to expect: "Walk me through an ML system you designed end-to-end, and where the offline-online gap surprised you." "How would you eval a recommendation system whose ground-truth feedback is delayed by 30 days?" "Your team's primary metric just plateaued for 6 months. How do you decide where to invest the next 4 quarters?" "Describe a time you mentored a junior engineer through a difficult promotion case." "How do you decide between fine-tuning an open model and using an API for a new capability?" Hello Interview's senior-level rubric (hellointerview.com/blog/understanding-job-levels-at-faang-companies) covers the cross-company patterns.

Compensation: the real bands at senior

Total comp at senior FAANG-tier and AI-labs in 2026 (US, per levels.fyi):

Company	Level	Base	Total comp
Meta DS	E5	$210k–$260k	$400k–$580k
Google MLE	L5	$220k–$270k	$420k–$600k
Netflix MLE	L5	$400k–$520k	$520k–$780k (single-band)
Anthropic Senior MTS	senior	$380k–$500k	$700k–$1.4M+
OpenAI Senior MTS	senior	$400k–$550k	$800k–$2M+ (heavy PPU)
Databricks MLE	L5	$240k–$310k	$420k–$680k
Scale AI	senior MLE	$280k–$360k	$500k–$900k
Hugging Face	senior	$210k–$270k (remote-friendly)	$340k–$540k

The structural fact at senior: AI-lab senior MTS comp at Anthropic and OpenAI commonly exceeds FAANG staff (L6/E6) comp on equity-heavy total. OpenAI's PPU has produced reported senior-MTS total comp in the $1.5M–$3M range during peak vesting in public levels.fyi reports. Risk-adjusted, the bands converge — AI-lab equity is more concentrated; FAANG comp is more diversified across years and stock cycles.

Frequently asked questions

What's the difference between senior DS and senior MLE in 2026?: Title shape varies by company; the day-to-day diverges. Senior DS at analytics-shape companies (Meta, Airbnb, Stripe, Netflix Studio, Uber Marketplace) is product-analytics-leaning: experiment design, causal inference, metric ownership, cross-functional partnership with PMs. Senior MLE at production-ML-shape companies (Google, Meta Ads, Netflix Recs, Apple Intelligence) is engineering-leaning: model training infra, deployment surfaces, monitoring, on-call. The two career paths converge again at staff+ where 'broad ML domain ownership' is the title regardless of analytics-vs-MLE history.
How important is on-call rotation at senior?: Significant at production-ML-shape companies. Senior MLEs at Netflix, Google, Meta Ads, and most large-scale production-ML teams are in the on-call rotation for their model-serving systems. The bar: ability to debug a p99-latency regression at 3am, identify whether it's a model issue or a serving-infrastructure issue, run the rollback playbook, write the post-incident document. Analytics-DS senior roles typically have a softer on-call shape — pager for data-pipeline failures or experiment-readout escalations, not 24/7 latency monitoring.
Should I move to an AI lab from FAANG at senior?: Depends on what you're optimizing for. AI labs (Anthropic, OpenAI, Scale AI, xAI, Cohere, Together) pay materially higher equity-heavy total comp at senior — commonly $700k–$1.4M+. The work is closer to research and frontier-model engineering. The risk: equity concentration in a single private company; less diversified than FAANG public stock. The career signal: AI-lab senior MTS is highly transferable to other AI labs and to staff-track roles at FAANG; a research-eng background opens doors that pure-application-ML doesn't. Worth the move for engineers who want frontier work and accept the equity-concentration risk.
How do I know I'm ready for staff promotion?: Three artifacts: (1) you've led one cross-team initiative whose outcome named you as load-bearing; (2) you've mentored at least one junior or mid engineer through a promotion under your sponsorship; (3) you have a documented strategic articulation — a one-pager or roadmap that ties your team's ML work to company-strategic outcomes. If all three are not in place 12 months before you go up for staff, you're not ready. Promotion takes 2–4 years from senior at most large tech companies.
What's the right approach to LLM / foundation-model work at senior?: Be conversational about the trade-off space, not the latest paper. The bar at senior in 2026: comfort fine-tuning open models at scale (FSDP / DeepSpeed Stage-3), opinion on prompt engineering vs PEFT vs full fine-tune for a given problem, fluency with the eval-harness ecosystem (lm-evaluation-harness, EleutherAI's framework at github.com/EleutherAI/lm-evaluation-harness), and an articulated view on RAG architectures (vector search vs dense retrieval vs graph retrieval). Reading every new arxiv paper is junior; having an opinion on which arxiv papers matter for your domain is senior.
Do I need to publish papers at senior?: Required at AI-lab research-track senior MTS. Optional at FAANG production-ML senior. Anthropic and OpenAI senior MTS roles often have a research-publication expectation; the public alignment-research and safety-research papers are co-authored by senior MTS. FAANG production-ML senior roles weight internal-impact (lift on the north-star metric) more than publications. Publishing one or two papers as a senior is a strong signal for staff-track promotion at FAANG and is increasingly expected at AI-labs.

Sources

About the author. Blake Crosley founded ResumeGeni and writes about data science, machine learning, hiring technology, and ATS optimization. More writing at blakecrosley.com.