The hiring profile: research lab, not consumer app
The company that once chased consumer virality now hires like a research lab. That shift — quiet, deliberate, nearly invisible from outside — decides which resumes clear the first screen.
Character.AI's open roles tell the story. As of September 2026, the board shows 13 salaried positions across engineering, research, data, and a single technical program management slot. The listings cluster around three titles: Principal Research Engineer, Post-Training; Research Engineer, AI Safety & Alignment; and Research Engineer, Post-Training (All Industry Levels). A Machine Learning Infrastructure Engineer role and a Staff Data Scientist, Monetization round out the technical core. These are not generalist "AI engineer" postings. Each targets a specific layer of the model lifecycle, post-training optimization, safety alignment, infrastructure scaling, or monetization analytics.
The function mix confirms the narrowing. Engineering holds 10 of 13 roles; research, data analytics, operations, and product claim one each. The consumer product team that once ballooned to support millions of roleplay users has contracted. A 5% layoff in August 2024 and the founders' return to Google DeepMind that same year marked the pivot. What remains is a team building for model quality, not user growth.
Seniority skews hard toward the top. EngRadar data shows 11 roles with unspecified seniority and 2 principal roles. The first-party board reveals the actual floor: Principal Research Engineer, Staff Data Scientist, Research Engineer (All Industry Levels). The "All Industry Levels" label on the Post-Training role is the only concession to breadth, and even that sits in a $225k–$400k band.
Teams map directly to the research agenda. Post-training — the RLHF, preference optimization, and evaluation loop — absorbs the largest share of engineering and research headcount. Safety & Alignment operates as a dedicated vertical. Infrastructure supports the compute demands of both. Monetization data science exists as a single staff role. Trust & safety appears as an operational function.
Geography reinforces the profile. Eleven of 11 tracked roles sit in the San Francisco Bay Area; New York appears as a secondary hub. Hybrid dominates at roughly three in four roles, but the onsite expectation for research-heavy positions is implicit in the Redwood City anchoring of every first-party listing.
This is not a company hiring for potential. It hires for demonstrated capability in the exact problems it faces today.
Compensation: what the bands show
Character.AI pays like a research lab that happens to be a consumer product company, senior-heavy, equity-rich, transparent enough that you can walk into negotiation with a printed band. Zero G Talent's board data puts a salary range of roughly $150k to $400k base, with a median around $300k across 13 active salaried roles. The range is wide because the company hires almost exclusively at staff-and-above levels; there is no junior ladder to compress the bottom.
| Role | Location | Base Salary Band (USD/year) |
|---|---|---|
| Principal Research Engineer, Post-Training | Redwood City, CA | $275,000 – $400,000 |
| Research Engineer, AI Safety & Alignment | Redwood City, CA | $225,000 – $400,000 |
| Research Engineer, Post-Training (All Industry Levels) | Redwood City, CA or New York City | $225,000 – $400,000 |
| Machine Learning Infrastructure Engineer | Redwood City, CA | $150,000 – $350,000 |
| Technical Program Manager, Product Experience | Redwood City, CA | $200,000 – $300,000 |
| Staff Data Scientist, Monetization | Redwood City, CA | $250,000 – $300,000 |
These bands come from live Character.AI postings on our board as of mid-2026. They reflect what the company puts in writing before a candidate applies, not a recruiter's verbal range.
Third-party aggregators tell a consistent story. Levels.fyi's data shows a median Software Engineer total compensation of $300k per year in the United States, with its highest-recorded role at Character.ai, a Recruiter at the Common Range Average level, hitting $382,500 total. Jobsbyculture.com's figures put the 2026 engineer total-comp window at $200k–$650k. The spread between median and ceiling is almost entirely equity.
Equity grants vest over four years on a standard monthly schedule with a one-year cliff. The company does not disclose strike prices or preferred valuations in offer letters; candidates who ask get the current 409A and the last primary round price.
Inside the interview loop
The Character.AI software engineer interview is a focused, fast-moving pipeline that typically runs three to five weeks from first contact to offer, generally faster than FAANG loops. Most candidates report a process that rewards systems thinking and a solid grasp of how large language models are actually served in production.
The funnel follows five stages: application and résumé screen handled by an ATS plus a recruiter eligibility check; an online assessment or take-home coding exercise calibrated to level; live technical interviews covering coding, design, and domain depth; a hiring-manager or bar-raiser round for level calibration and team fit; offer negotiation and background checks. U.S.-facing loops for mid-level roles often span two to six weeks; senior loops can run longer when panels and compensation approvals stack.
The first human contact is a 30-minute recruiter screen. The hiring-manager screen follows at 45 minutes. The technical phone screen runs 60 minutes: one coding problem plus a discussion of ML inference or distributed systems. Coding questions are standard LeetCode mediums, Python-flavored for most roles. Interviewers expect two problems in a 45-minute window across arrays, strings, hash maps, and the occasional graph or heap; talking through your approach and edge cases counts as much as landing a working solution.
Dataford's analysis of 11 interview reports shows the topic coverage: React 100%, deep versus shallow copy 96%, maps and hash maps 92%, algorithms 89%, timestamped data and temporal queries 85%, data-structures planning 78%, code quality and communication during coding 75%, time-complexity optimization 71%, custom data-structure implementation 67%, front-end systems design 64%, React state and re-render behavior 60%, Python 60%.
The ML inference discussion separates candidates who have read papers from those who have run models in production. Interviewers probe KV-cache management, batching strategies, GPU memory pressure, and speculative decoding. They want to see whether you understand that memory, not raw compute, usually caps how many conversations one GPU can hold, so be ready to reason about tokens-per-second and cost-per-conversation rather than correctness alone.
The virtual onsite spans four to five rounds. System-design prompts are specific to the product: design a high-throughput chat with persistent conversation state, or design a content-moderation pipeline for AI outputs. Difficulty distribution across reports skews medium (55%), with easy at 27% and hard at 18%; overall difficulty rates 4.8 out of 10. Experience sentiment is 64% positive, 27% neutral, 9% negative.
Candidates disqualify themselves by failing to articulate the memory-bounded nature of LLM serving. Treating the coding round as a silent LeetCode grind, communication and edge-case reasoning are explicitly scored. The bar is research-adjacent: the team hires engineers who can move between a PyTorch checkpoint and a production Kubernetes service without hand-off friction.
The workplace: Menlo Park anchor, New York satellite
Character.AI's physical footprint centers on Menlo Park, where the company has anchored its headquarters since founding in 2021. The address appears across sources as 800 W El Camino Real, Menlo Park, CA 94025, with CB Insights listing a nearby suite at 700 El Camino Real #1152, Suite 120 (likely the same campus or adjacent buildings in the Stanford Research Park corridor). Salestools.io describes the site as the main operational hub housing engineering, product, sales, and executive leadership, and the company's own job board confirms the concentration: every first-party listing anchors in Redwood City, the neighboring city that shares the El Camino Real artery and effectively functions as a single campus zone. One role, Research Engineer, Post-Training (All Industry Levels), adds New York City as an alternative, signaling a second, smaller hub.
Built In SF reports two total offices, HQ Palo Alto and Menlo Park, plus a "Remote Workspace" designation, while Salestools.io cites 200+ employees worldwide. The discrepancy with Built In's 30-employee figure suggests the latter may reflect an older snapshot or a narrower definition of "office-based" headcount. CB Insights classifies the company's stage as "Reverse Acqui-Hire | Alive" as of September 2026.
The Menlo Park workspace follows a modern hybrid template: open-plan floors with flexible seating, designated quiet zones for deep work, conference rooms and team meeting areas, high-speed connectivity and advanced systems, fitness centers and wellness rooms, cafeterias and beverage stations, and lounges and social spaces. Salestools.io frames these as deliberate culture carriers, "thoughtfully designed workspaces that encourage both focused work and dynamic team collaboration", and the amenity mix aligns with the JLL workplace principles the research cites: a consistent baseline (standard desk sizes, screen specs, chair quality, meeting-room tech) layered with inclusive design (wellness rooms, multi-faith or quiet-reflection spaces) and local character. Companies at this scale typically deliver inclusive amenities themselves because landlords rarely embed them in base builds.
Remote work is an option, but the concentration of senior research postings in Redwood City signals that the core technical work still orbits the physical cluster around El Camino Real.
Who lasts here
Character.AI runs lean. BuiltIn lists the headcount at roughly 30 people across two offices, Palo Alto headquarters and a New York presence, yet the roles on our board span principal research, post-training, safety and alignment, ML infrastructure, and a staff data-science seat for monetization.
The product itself selects for a specific temperament. Character.AI's users — predominantly female, per the Washington Post's December 2024 reporting — spend hours building relationships with anime and gaming characters. Reddit threads from 2023 capture the user side: "some people that lack the people and social skills finally have an outlet to talk to something," and "some people use it to simulate relationships with a character."
Research from the AI-companion ecosystem underscores the technical bar. The Luvr.ai 2024 guide notes that "consistency is key for believable AI personalities" but "it's also important to let characters grow a bit to keep chats interesting over time," and recommends mixing MBTI, Big Five, and Enneagram frameworks to model decision-making and motivation.
Stanford GSB's 2023 study on CEO personality and culture found that extraverted, sociable leaders correlate with agility and collaboration, while highly conscientious, detail-oriented leaders correlate with less innovation and, surprisingly, less execution. Deloitte's 2024 State of AI in the Enterprise survey found high-achieving AI organizations report more fear than low achievers, not because they're risk-averse, but because they're close enough to the frontier to see the cliffs.
The board's Technical Program Manager role ($200k–$300k) sits at the intersection: translating research milestones into product releases without slowing the research. Same for the Staff Data Scientist, Monetization ($250k–$300k): the business model is still forming, so the analyst defines the questions, not just the queries.
Character.AI hires the person who already did it.
Working in AI? Zero G Talent tracks the openings: see every open Character.AI role, browse AI jobs, the companies hiring, and the people building the field.