Companies — tier list & profiles
Filtered for your stack: Bay Area only · ML-focused · top tier · Sr Staff/Staff IC · no finance. Ranked specifically for a RecSys + pretraining background.
Filter rules applied
- Geography: SF / Peninsula / South Bay / East Bay only. Dropped: Skild (Pittsburgh), Runway (NY), Modal (NY), General Intuition (NY), Liquid AI (Cambridge MA), Reflection (Brooklyn), Anduril (Costa Mesa primary), Snap (Santa Monica).
- No finance: dropped all trading firms (Renaissance, TGS, XTX, Spark, Radix, HRT, Jane Street, Citadel, PDT, D.E. Shaw, Jump, Optiver, Two Sigma, Cubist, Verition, Millennium). Coinbase / Robinhood / Plaid kept as watch-only.
- ML-relevant: must have a real ML org and Staff+ ML/RE/RS roles. Dropped Aetherflux (space power, not ML), Disney (per your call), Wayfair, Bloomberg, Atlassian's non-AI teams.
- Top tier: median Staff TC ≥ ~$400k OR equity bet worth taking (early frontier).
The tier list (S → C)
Tier S — Frontier labs (apply first, with referrals)
Anthropic S
Comp: Staff $920k · T6 Tech Lead RS $2.0–2.5m · monthly vest after 1y cliff. Coatue 2030 estimate $2T.
Roles to target: MTS Pretraining, MTS Post-Training (RLHF/RLAIF), MTS Inference Performance, Product Eng (Claude apps).
Loop: ~20 days. CodeSignal take-home (4-level progressive spec, 90 min). Then onsite: ML round (Colab Q6), concurrency-flavored coding, ML config system design, behavioral. Reference checks during the loop.
Why for you: pretraining + post-training are the strongest teams. RecSys experience translates to personalization / product surface (Memory, Projects).
Levers: Mike Krieger (CPO) sources senior product IC directly via LinkedIn. Best: warm intro from MTS-level (Tom Brown / Sam McCandlish / Jared Kaplan area).
OpenAI S
Comp: L5 Staff ~$875k–1.5m · L6 Sr Staff $1.2–2.0m+. RSU (switched from PPU for new hires Jan 2026), immediate quarterly vest. Highest comp band on this list.
Roles to target: MTS Pretraining, MTS Post-Training, MTS Personalization/Memory, MTS Search/Embeddings, Applied AI Eng.
Loop: 2–8 weeks. CoderPad multi-part problems (60–75 min each). 4-6 hr final loop in 1–2 days. Coding bar is binary — they reject for 2/4 even if ML/research/behavioral nail it.
Why for you: Personalization / Memory team is your direct pivot — RecSys at GPT scale. Pretraining team is the alternative.
Levers: many ex-Meta in Pretraining + Multimodal — easy network bridge. Look for ex-Meta GenAI / RecSys colleagues now there.
Thinking Machines S
Comp: $350k–475k base (filings); SWE/RE/RS tracks across Data Infra, Inference, Kernels, RL Systems, Tinker, Pretraining Datasets/Science, Post-Training, Synthetic Data, Vision, Audio. Equity at $50B → $200B+ trajectory.
Roles to target: MTS Pretraining, MTS RL Systems, MTS Inference, MTS Post-Training. Also product eng for Tinker.
Loop: Glassdoor n=14. Reportedly research-style + RE coding, OpenAI/Anthropic-tier bar. Multi-round, ~19 days.
Why for you: Andrew Tulloch is ex-Meta and reportedly turned down $1.5B from Meta to be there. They want infra+modeling depth. Tiny team = high impact.
Levers: ex-Meta GenAI alumni network. Read Mira / John / Lilian's public writing before the loop.
See: TML prep deep dive
Safe Superintelligence (SSI) S
Comp: not public. Likely >$700k base just to attract from FAANG.
Roles: all "Technical Staff" — no public job board. Apply via website, slow.
Loop: famously secretive. Vetting candidates for "good character" over hours.
Why for you: Ilya-led pretraining + safety. Long shot but huge upside.
Levers: Daniel Gross (Andromeda) or Daniel Levy.
Meta MSL Sinternal
If you are eligible to apply internally to Meta Superintelligence Labs (the TBD Lab Zuck/Wang are recruiting for), this is the highest-comp option on Earth right now.
Catch: external hires only via Zuck/Wang direct. Internal transfers possible but very competitive — must be top-of-band performer.
Action: ping your skip / VP about MSL fit before you announce externally. Once you're seen as a flight risk, it's harder to negotiate the internal move.
Tier A — Strong Bay Area AI labs (apply weeks 2–3)
Google DeepMind A
Comp: L6 Staff $700k+ · L7 Sr Staff $950k–1.1m. With DeepMind premium and Gemini budget, top of band $700k–$1m for verified pretraining experts.
Roles: Sr Staff Research Engineer (Gemini pre/post-training), Staff SWE on Gemini, Staff MLE on YouTube RecSys.
Loop: 41-day average. Paper deep-dive (60 min) + research framing (60 min) + ML coding (60 min, no AI tools) + standard Google loop. PhD-defense-grade depth.
Why for you: Gemini pretraining + YouTube RecSys are direct adjacencies. Lots of ex-Meta in YouTube ML.
xAI A
Comp: ~$1.2m Staff MTS · immediate monthly vest. SpaceX merger means equity is closer-to-liquid than other private labs.
Roles: SWE (frontier), ML Research Engineer, RecSys (X integration).
Loop: 2–4 weeks (faster than peers). 20-min "hardest tech problem" presentation + Q&A defense. CodeSignal OA (3 problems, 60 min). Mix of algos and thread-safety/concurrency. "Simple and correct beats clever and broken."
Why for you: Colossus pretraining at 200K+ H100s. Recsys for X feed.
Watch: Musk culture + intense PA in-person. Mixed Blind reports. SpaceX merger means equity is closer-to-liquid (IPO mid-2026 per tier list).
Cursor A
Comp: median $1.1m · range $850k–1.28m+. Highest median software comp on Levels.fyi. Tracks: Client Infra, Data Infra, Enterprise, Growth, Infrastructure, ML Infrastructure, Product Eng, RL Scaling, Research Scientist, Security.
Roles: MTS Models, MTS Infra, MTS Tab. Tiny team, extreme bar, ~1/50 candidates pass.
Loop: TypeScript-first (Python ok for ML roles). AI tools permitted but interviewer judges your prompts. Sr/Staff: 4–8 hr take-home in their actual codebase. Streaming Markdown parser, streaming edit application, multi-file diff coordination.
Why for you: Tab models are pretrained code models — pretraining experience direct fit.
Culture: flat, no PMs, in-person SF, 50+ hr weeks. Long-shot but stratospheric upside.
Physical Intelligence (π) A
Robotics foundation models (π0 series). Pretraining for embodied agents. Modeling-heavy.
Comp: likely $500k–900k. Roles: ML Research Engineer, Foundation Model Eng. Hot space, major upside.
World Labs A
Fei-Fei Li-led. Spatial AI / world models. Pretraining + multimodal modeling.
Roles: Computer Vision, Graphics, Robotics-intersection ML.
Mistral (Palo Alto US hub) A
Pretraining lab. PA US hub focused on Applied AI + Research Engineer hiring. Smaller US team (~50), less RTO-strict.
Comp: US PA hires reportedly $400k–700k total. Equity upside lower (European centroid).
Loop: ~7 stages — LLM theory → coding → past project → tech manager → ML system design → take-home → values.
Periodic Labs A
AI scientist for materials science. Foundation models on scientific data — pretraining-adjacent. Founders ex-OpenAI/DeepMind, backed by Jeff Dean / Eric Schmidt / Bezos.
Reflection A
Frontier model lab. Pretraining + autonomous coding agents. Aims for Western open-source frontier model competing with advanced China systems. Brooklyn HQ — fails your geo filter. Watch only unless they open a Bay Area office.
Magic.dev A
Code model frontier lab with massive in-house compute. Hiring MTS Kernels, Supercomputing Platform. Comp likely $600k–900k+. Pretraining fit if confirmed by recruiter.
Mirendil A
Ex-Anthropic researchers (Behnam Neyshabur — pretraining theory). Bio + materials science. Strong pretraining DNA, niche domain. Watch for public Staff postings.
Tier B — Strong Bay Area, RecSys-direct fallback
| Company | HQ | Staff TC | Why for you | Loop notes |
|---|---|---|---|---|
| SF | $570k–959k (IC17) | Pure RecSys at scale; ex-Meta network strong | 3 ML fundamentals + 2 LC hard tech screen; ML theory + LC mediums onsite. Difficulty 3.4/5 | |
| Roblox | San Mateo | $750k | Pure RecSys, gaming context | HackerRank OA + Roblox Assessment + 2 onsite (med + hard) + behavioral |
| SF | $610k | Feed RecSys + ads. Fully remote OK. | Tech screen (build a model from data) → onsite (model design, feature eng, ad applied ML) | |
| Uber | SF | $620k–723k median | RecSys + ranking, scale | 5–7 rounds over 4–8 weeks; 60/40 ML/coding for MLE; recent DP-on-trees screening reports 2025 |
| Airbnb | SF | $548k median, up to $1.64m | Trust signals, dynamic pricing, personalization | LC med-hard, Airbnb-tagged. DP problems on trees common. |
| DoorDash | SF | $550k | Geospatial, time-varying ranking, supply-aware | 60-min DSA round; LC medium with optimize follow-up. Design HashMap, Jump Game |
| Instacart | SF | $545k | Same flavor as DoorDash | Bar Raiser; reference checks |
| Lyft | SF | $475k | Lower ceiling than Uber but RecSys/eta similar | Standard FAANG-style |
| AppLovin | Palo Alto | $600k | Heavy ML/recsys for ad targeting; pure RecSys fit | 5 rounds: 3 tech + 1 sys design + 1 behavioral. Classic LRU O(1) → real-system extension. ~3 weeks total. |
| Mountain View | $402k–700k | Feed/jobs RecSys at scale; MS RSU | Standard FAANG loop | |
| Databricks | SF | $504k–800k–1.65m (L7) | Mosaic ML pretraining infra | 1hr CoderPad screen, LC med-hard, ~20% pass. Reference checks (3). |
| Nvidia | Santa Clara | $450k–675k | RecSys, robotics ML, inference. Comp lower than frontier. | ML system design + project depth. Stars pyramid, ring buffer, parallel coding. |
| Apple ML / GenAI | Cupertino | $550k–865k | Rebuild mode post-Ruoming-Pang exit; senior pretraining hires getting $1m+ | 5–8 rounds × 45–60 min. 1–2 LC med-hard (rotate matrix, merge K, LRU). Some ML+DSA blends. |
| Sierra | SF | $15.8B startup | Customer service AI agents. Bret Taylor co-founder. | Replaced LeetCode with AI-native onsite: planning + 2-hr building phase using AI tools. |
| Cohere | SF/Toronto | ~$700k | Pretraining + enterprise post-training | Recruiter → HM (projects) → onsite: coding + ML + sys design + behavioral. Forward-deployed-platform-engineer flavor. |
| Perplexity | SF | $450k median, up to $790k | Search + LLM. RecSys partial fit. | Fast ~23-day loop. Python only. Streaming dedup, top-k, beam search, batch inference optimization. |
| Decagon | SF | $4.5B startup | Voice/customer agents | Coding pair + sys design + past project + behavioral. Tic-Tac-Toe, conversation rolling-window scores, LC #84. |
| Cognition (Devin) | SF | $10.2B startup | Coding agents | Customer-facing roleplay + pair programming + architecture. "Customer is angry" simulations. |
| Glean | Palo Alto | $7.25B | Enterprise search + agents | MLE for AI Assistant + Agents |
Tier B continued — Inference / hardware infra (good fit if you want infra pivot)
| Company | HQ | Notes |
|---|---|---|
| Together AI | SF | $3.3B. Open-source inference + post-training infra. |
| Fireworks AI | Redwood City | $4B. LLM inference. |
| Baseten | SF | $5B. Inference platform. |
| Cerebras | Sunnyvale | $8.1B. Wafer-scale chips. ~26-day loop, LC med + parallel programming + matmul. |
| Groq | Mountain View | $6.9B. LPU inference. |
| Modular | Los Altos | $1.6B. Mojo + AI inference engine. Chris Lattner. |
| MatX | Mountain View | ASICs for LLMs. Ex-Google TPU team. |
Strategy: which tier to attack when
- Week 1–2 (warmup): apply to Tier B first (Pinterest, Roblox, Databricks, Cohere). Use these onsites as paid practice. Expect to bomb at least one.
- Week 3–4: apply to Tier A (DeepMind, xAI, Cursor, PI, World Labs, Mistral, Periodic, Magic).
- Week 5–6: apply to Tier S (Anthropic, OpenAI, TML, SSI). By now you'll be sharp from earlier loops.
- Always: keep at least 2 active Tier B in your funnel as offer leverage.
Companies dropped (and why)
| Company | Reason |
|---|---|
| Skild AI | Pittsburgh HQ, no Bay Area |
| Runway | NY HQ |
| Modal | NY HQ |
| General Intuition | NY HQ |
| Liquid AI | Cambridge MA |
| Reflection (Brooklyn $130m round) | Brooklyn (different from $2B SF lab) |
| Anduril | Costa Mesa primary; small Bay Area presence — Watch only |
| Snap | Santa Monica (LA) — bordeline; drop on geo |
| Crusoe | Denver HQ (SF is sales hub) — Watch |
| Surge AI | data labeling, no Staff modeling roles — Watch C |
| Aetherflux | space-power company, not ML |
| All trading firms | finance filter |
| Disney | per your call |