Who gets hired and onto which teams
FundedIQ reported that Mirelo AI closed a $41 million seed round in December 2025, co-led by Andreessen Horowitz and Index Ventures, to build generative audio models that turn video into sound effects, music, and synced audio. The company is hiring engineers who can ship production-grade models with minimal supervision. The Berlin-based lab operates remote-first, and Zero G Talent's board data shows its eight open roles span every layer of the stack.
Co-founders Florian Wenzel and Carl Johann Simon-Gabriel spent roughly a decade in AI research at the AWS Lab on large vision models and later large language models before founding Mirelo in 2023. That background shaped a hiring philosophy that prizes systems engineering ability. FundedIQ's figures put the team in the 11-to-50 employee band as of the seed round. Simon-Gabriel said the goal is to at least double, if not triple, the model team's size with the new capital.
The model team forms the core. Two research-scientist postings — Research Scientist - Model Team and Research Scientist - Audio Codec, both based in Berlin — reflect the architectural split the co-founders described: a music model and a video-to-sound-effect model, each dependent on efficient audio tokenization.
Training infrastructure gets its own dedicated hire. The Full-Stack Software Engineer (Berlin) will ship end-to-end features across Mirelo Studio and the API that powers the Model Context Protocol server, the hosted MCP server that lets AI assistants call sound-generation tools (Generate from text, Generate from video, Extend, Inpaint) directly from an IDE. The Principal Product Designer and Principal Product Manager (both hybrid) carry "Principal" scope.
Go-to-market roles mirror the two revenue streams. The Founder Associate (hybrid) and Community and Creator Growth Manager (Berlin) split across B2C studio adoption and B2B partner integration. Simon-Gabriel described the need for marketing, growth, and sales capacity to handle both motions simultaneously. The hiring signal is consistent across every role: remote-first (Berlin or hybrid), equity, and a bar set at demonstrable systems engineering ability. The next section breaks down what that compensation structure looks like in practice.
What it pays
No Mirelo-specific salary bands have been disclosed publicly. The company's job posts state only "competitive compensation and equity" with "strong packages that ensure you share in the success you help create." To calibrate expectations, the nearest usable benchmarks are summarized below.
| Source / Role | Base Salary Range | Total Compensation | Details |
|---|---|---|---|
| Remote AI Engineers (26 listings, 2026) | $154K–$209K (typical) | — | Floor $60K, ceiling $325K |
| Remote AI Engineers — Senior (18 data points) | $162K–$215K | — | — |
| Remote AI Engineers — All-level (8 data points) | $135K–$199K | — | Global remote medians |
| Hyperscalers — Senior AI Engineers | $200K–$230K (median ~$215K) | — | Tight band |
| Meta E5 | — | $480K | — |
| OpenAI | $336K | $1.1M | Split: $336K base + $770K profit units |
| Production ML Infrastructure (premium) | +$50K+ vs generalist ML | — | At same level |
| Netflix — Senior Engineers | $400K–$600K+ | — | All-cash |
These figures are global remote medians or public data points, not Mirelo offers. Market context matters. Amazon back-loads equity 5/15/40/40. OpenAI's profit units lack public-market liquidity. Mirelo, as a seed-stage German company hiring hybrid in Berlin and Tübingen, will not match those absolute numbers.
What a seed-stage AI lab typically offers: base salaries calibrated to European tech markets, meaningful equity grants, and milestone-tied refreshes. The Training Infrastructure Engineer and research-scientist roles, which demand production-grade distributed training and audio-codec expertise, sit at the top of the internal band. The Full-Stack Software Engineer and Principal Product Manager roles likely bracket the middle. The Founder Associate and the growth manager role carry more variable, growth-linked upside.
Equity at this stage is paper until a liquidity event. Vesting schedules, cliff terms, and refresh policies have not been published. Candidates should ask directly: current strike price, total shares outstanding, and whether the company uses standard four-year monthly vesting or a non-standard schedule. Remote workers outside Germany should also confirm whether compensation is localized: the market data shows up to a 15% discount versus Bay Area rates for fully remote AI roles.
How the hiring process works and what gets candidates through it
Mirelo AI's hiring timeline is notably compressed. Glassdoor data shows an average of 14 days from application to offer across all roles, based on one user-submitted interview, a figure that aligns with the lean, research-heavy structure visible on the company's AshbyHQ portal. The same source lists three distinct interview questions and four anonymous reviews on the U.S. Glassdoor site, plus two questions and three reviews on the U.K./India portal, suggesting a consistent but lightweight evaluation loop rather than a multi-month gauntlet.
The process follows a standard research-lab arc: an initial screen with a recruiter or hiring manager, a technical deep-dive, and a final round that blends research discussion with team fit. Candidates for the research-scientist roles should expect questions that probe PyTorch internals, distributed training trade-offs, and the ability to ship a model from experiment to production without hand-holding. The Training Infrastructure Engineer opening signals that the bar includes cluster-level thinking: GPU utilization, checkpointing strategies, and the plumbing that keeps a large-scale run from stalling.
Screening criteria, while not published verbatim, can be read from the role descriptions on the Mirelo AI board. Every research and engineering posting emphasizes "production-grade" delivery, "minimal supervision," and "systems engineering ability," phrasing that mirrors the company's stated preference for demonstrable shipping. For the role, the ask shifts to end-to-end product sense: deploying inference APIs, building evaluation dashboards, and owning the latency budget from model to user.
Application strengths that convert tend to cluster around three signals. First, a portfolio artifact — a trained model, a kernel optimization, or a data pipeline — that the candidate can walk through, explaining the failure modes they hit and how they fixed them. Second, evidence of cross-stack fluency: researchers who have written their own data loaders, engineers who have debugged a divergence in a large-parameter run. Third, clear communication in async formats; the hybrid setup means written design docs and Loom walkthroughs are daily currency.
Where the work happens
Mirelo AI operates from a research lab in Berlin, the city where the company is commercially registered through the Amtsgericht Charlottenburg. The legal entity, Mirelo AI GmbH, lists its address at Eisenbahnstraße 1 in Tübingen, but the team's day-to-day work centers on the Berlin facility described on the company's about page as "our lab in Berlin, a small but exceptional team is creating what comes next." That lab hosts the research scientists, infrastructure engineers, and full-stack engineers who build the audio models and the APIs that serve them.
The Berlin location is not a generic coworking space. The work there requires GPU clusters for training frontier audio models (SFX 2.0 and the Audio-to-MIDI Pro system) and the engineering infrastructure to serve those models through the Mirelo API, on Runway, on fal, and through the Kiro Power integration. The Training Infrastructure Engineer role posted on the Zero G Talent board is based in Berlin, as are both Research Scientist positions (Audio Codec and Model Team) and the Full-Stack Software Engineer role. Those postings signal that the lab contains the compute, storage, and networking needed to iterate on large generative audio models and ship them to production endpoints.
Hybrid roles (Founder Associate, Principal Product Manager, Principal Product Designer) indicate the company supports distributed work for functions that don't require daily physical access to the training cluster. The Community and Creator Growth Manager role is also listed in Berlin, suggesting that even go-to-market roles stay close to the product team.
The AWS connection is concrete. Mirelo's LinkedIn posts note that "AWS published the full story of how we built it," referring to the Kiro Power integration that lets developers generate sound effects inside their IDE. That publication implies the lab's training and serving stack runs on AWS infrastructure, or at least that AWS was a key partner in the build.
Tübingen appears in the imprint as the registered address. The commercial register entry (HRB 258001 B) confirms the Berlin jurisdiction. For a candidate, the distinction matters for tax and visa paperwork; the work happens in Berlin.
What enables the team's work is the combination of a dedicated GPU-equipped lab in a city with deep ML talent pools (Max Planck, ETH Zurich alumni are named in the founders' backgrounds), a cloud partnership that removes infrastructure ceiling, and a hybrid policy that lets the company recruit product and design talent beyond commuting distance. The board's current openings (seven of eight roles tied to Berlin) show the lab is the center of gravity.
Who thrives here
The evidence shows a team that ships production-grade audio models at a pace unusual for a lab of this size. SFX 2.0 launched with blind-test wins over Sony's Woosh (81% preference) and over the native audio from Veo, Seedance, and Kling. Audio-to-MIDI Pro reached nearly 1 in 11 active users on riffle within six weeks. The Kiro Powers integration with AWS shipped with a published build story. These outcomes point to a handful of traits that correlate with success and retention at Mirelo.
Research depth that survives contact with product. The founders (Carl Johann and Florian Wenzel) came from AWS Labs, Google Brain, Max Planck, and ETH Zurich. The board lists two distinct research-scientist openings: Research Scientist - Audio Codec and Research Scientist - Model Team, both in Berlin. That split signals an expectation that researchers own a substantive slice of the model stack (codec or core modeling) and take it to launch.
Systems engineering as a first-class discipline. The Training Infrastructure Engineer role sits alongside the research roles. SFX 2.0's consistency guarantee — "hold the same sound across every cut for infinite length, so the same car sounds the same from your first shot to your last" — is a systems promise as much as a model promise.
Product judgment grounded in creative workflows. The Principal Product Manager and Principal Product Designer roles are hybrid (Berlin or remote), not pure research. The product surface is creative professionals: video makers who need "effects and ambience that match the picture," musicians who need "editable MIDI for every instrument," developers who type "Swooshes for a game" into Kiro and expect a synced file in their project.
Cross-disciplinary fluency without title inflation. The founders themselves embody the blend: conservatory-trained pianist and electro guitarist, plus top-tier ML credentials. The growth manager role exists because adoption lives in creator communities, not in benchmark leaderboards. The Founder Associate role is hybrid and explicitly high-leverage.
Autonomy with a shipping cadence. Remote-first and hybrid listings dominate the board. The lab's own about page says the team is building what's next. The trade-off is visible accountability: blind listening tests, public API uptime, AWS-published architecture posts.
Competitive calibration against the best public baselines. The 81% win rate over Sony's Woosh wasn't an internal metric; it was a blind listening test published on LinkedIn. The Audio-to-MIDI Pro launch called out "biggest gain on drums" and fixed timeouts that had blocked users earlier in the year. The team benchmarks against shipped competitors (Veo, Seedance, Kling, Sony) and publishes the results.
Glassdoor lists three interview questions and four interview reviews for Mirelo AI, plus employee reviews on salaries and culture; but the public content is thin enough that any trait list derived solely from it would be speculative. The stronger signal is in what the company ships, how it structures its roles, and who its founders and backers are. The pattern that emerges: researchers who think in systems, engineers who think in product, product people who think in creative workflows, and all of them comfortable owning a metric that the market can verify.
The $41 million round bought Mirelo the compute to train large audio models and the runway to prove a small team can out-ship labs with far larger headcounts. The next blind listening test will be public, too.
Working in frontier tech? Zero G Talent tracks the openings: see every open Mirelo AI role, browse frontier tech jobs, the companies hiring, and the people building the field.