Software Engineer,Multimedia & Multimodal AI Job Lansing area,Michigan USA,Software Development

Summary:

Applied AI (AAI) is Meta’s organization focused on making our AI models best-in-class, starting with coding. Within AAI, the Multimedia & Multi Modality team covers the multimedia domain across every modality, on both the input and the output side of a model: image, video, audio, speech and music. We work directly with research, model-training and engineering partners across MSL, TBD and FAIR.

Current problems include evaluating video experiences, diagnosing multimedia model behavior, producing domain-expert agent tasks, and building the data and measurement pipelines multimodal capabilities are trained and judged against.

About the role

You will take a modality or a capability area, decide what data is worth producing and how it should be measured, and carry it from an open question through to a pipeline that runs and a measurement the org relies on. This is a multimodal role, not a text-only role. You will work across image, video, audio and speech, as model inputs and as model outputs, and the data and evaluations you own will cover media, not text alone.

You will choose where the pod invests, own outcomes beyond your individual contribution, set standards other engineers build against, and raise quality without becoming the review bottleneck.

Required Skills:

Software Engineer, Multimedia & Multimodal AI Responsibilities:

Design and build agentic workflows and pipelines, including human-in-the-loop and expert-in-the-loop designs, to automate data production and scale output past what manual authoring supports.

Design and own data pipelines at scale: ingestion, filtering, pseudo-labeling and captioning with attribute classifiers, and provenance tracking for audio corpora.

Build evaluation infrastructure: objective metrics (speaker/style similarity, codec and generator quality), human listening-test pipelines, and the correlation analysis that ties the two together.

Improve training efficiency and reliability — distributed training, GPU utilization, codec and tokenizer retraining, experiment management.

Reproduce and extend state‑of‑the‑art research: implement new methods from papers into our codebases and run rigorous ablations.

Mentor engineers on the team, contribute to hiring and onboarding, and raise the bar on evaluation and quality practice.

Build and train generative and representation models for speech, sound, and music — including text-, audio-, and video-conditioned generation, infilling, editing, and style transfer.

Minimum Qualifications:

Bachelor's degree in Computer Science, Computer Engineering, relevant technical field, or equivalent practical experience

6+ years of programming experience in a relevant language or 3+ years of experience + PhD

3+ years building ML systems in production or research settings

strong Python and Py Torch

Demonstrated experience with speech, audio, or music ML - ASR, TTS, audio codecs, music information retrieval, self-supervised audio representation learning, or audio generative modeling

Experience with large-scale data pipelines and distributed training

Track record of translating research ideas into working, measurable systems

Preferred Qualifications:

Publications at top venues (ICASSP, Interspeech, ISMIR, NeurIPS, ICML, ICLR) in speech, audio, or music

Generative modeling of continuous data (diffusion / flow matching, audio or vision), and demonstrated ability to switch domains and ramp quickly

Audio DSP depth — pitch detection, FFT, real‑time signal processing

Experience with disentangled or controllable generation (voice, emotion, style, instrumentation)

Experience building evaluation harnesses and human‑eval pipelines for generative audio

Music domain expertise: stem separation, mixing, lyrics/vocal conditioning

Experience designing benchmarks…