---
title: "Senior Inference Engineer, Real-Time Diffusion Transformers"
slug: "senior-inference-engineer-real-time-diffusion"
id: "61b70d66-76b2-40c4-aaf1-90e69e8a54cf"
department: "AI Infrastructure"
department_code: "INF"
location: "London, UK"
workplace_type: "Hybrid"
employment_type: "Full-time"
compensation: "£105k – £150k (annual package)"
is_remote: false
is_featured: false
headcount: 1
canonical_url: "https://jobs.talent247.co/jobs/senior-inference-engineer-real-time-diffusion"
markdown_url: "https://jobs.talent247.co/jobs/senior-inference-engineer-real-time-diffusion.md"
apply_endpoint: "https://jobs.talent247.co/api/public/jobs/senior-inference-engineer-real-time-diffusion/apply"
schema_endpoint: "https://jobs.talent247.co/api/public/jobs/senior-inference-engineer-real-time-diffusion/form-schema"
webmcp_tool: "apply_for_role"
updated_at: "2026-11-15T17:42:54.883Z"
content_signal: "search=yes, ai-input=yes, ai-train=no"
---

# Senior Inference Engineer, Real-Time Diffusion Transformers

## Position Telemetry

- **Reference:** `SENIOR-INFERENCE-E`
- **Practice / Sector:** AI Infrastructure (`INF`)
- **Location:** London, UK
- **Workplace Type:** Hybrid
- **Employment Type:** Full-time
- **Compensation:** £105k – £150k (annual package)
- **Canonical HTML URL:** https://jobs.talent247.co/jobs/senior-inference-engineer-real-time-diffusion

## Role Dossier & Specification

THE CLIENT & MANDATE
Our client is Britain's flagship $4B generative AI video unicorn headquartered in central London, serving more than 60,000 businesses worldwide with expressive neural video avatars and studio-grade generative video models. As their product evolves from asynchronous batch video rendering to live, conversational real-time video avatars that see, listen, react, and speak with sub-second latency, runtime inference efficiency on Diffusion Transformers (DiTs) has become a mission-critical engineering pillar.

They have retained Talent247 to appoint a Senior Inference Engineer, Real-Time Diffusion Transformers in London. Bridging their Generative Video Research group and Production Inference Infrastructure squad, the appointed engineer will take multi-billion-parameter causal audio-conditioned video diffusion and autoregressive models and make them stream infinitely long 1080p+ video sequences faster than real time on GPU clusters.

KEY RESPONSIBILITIES
• Profile, optimize, and deploy causal streaming Diffusion Transformer (DiT) and 3D spatiotemporal VAE architectures to achieve sustained >30 FPS real-time interactive avatar generation at <300ms time-to-first-frame.
• Implement custom CUDA, Triton, and TensorRT optimizations—including FP8/INT8/NVFP4 quantization, KV-cache rolling windows for causal video attention, CUDA Graphs, and fused audio-conditioning cross-attention kernels.
• Design multi-GPU sequence/context parallelism (Ulysses / Ring Attention) and disaggregated VAE decoding pipelines scaling smoothly across H100, H200, and B200 accelerator pods.
• Partner directly with research scientists on few-step adversarial/consistency distillation and causal architectural modifications that unlock 5x–10x runtime speedups without lip-sync or identity degradation.
• Integrate ultra-low-latency GPU video encoding (NVENC) directly with WebRTC media servers for zero-copy GPU-to-network frame streaming.
• Build automated perceptual-regression and frame-pacing benchmark suites guarding every runtime optimization.

TECHNICAL ENVIRONMENT & OFFERING
Their stack features Python, C++20, PyTorch (`torch.compile`), CUDA/CUTLASS, Triton, TensorRT, NVENC/WebRTC, and Kubernetes H100/B200 GPU fleets. Offered hybrid in central London, compensation includes £105,000–£150,000 GBP base salary plus Series E unicorn equity, comprehensive private healthcare, pension, and UK Skilled Worker / Global Talent visa sponsorship.

## Core Requirements

ESSENTIAL QUALIFICATIONS
• 5+ years of experience in GPU performance engineering, deep learning inference optimization, or high-performance multimedia/computer-vision systems.
• Deep hands-on mastery of PyTorch internals, CUDA/Triton kernel development, TensorRT, and hardware-level profiling using NVIDIA Nsight Systems and Nsight Compute.
• Direct practical experience optimizing transformer or diffusion architectures (DiTs, latent video diffusion, 3D VAEs, or LLMs) via FP8/INT8 quantization, operator fusion, and multi-GPU parallelism.
• Strong systems understanding of GPU memory hierarchies, PCIe/NVLink bandwidth, asynchronous CUDA streams, and tail-latency elimination.
• Collaborative mindset excited to co-design hardware-efficient architectures side-by-side with generative video researchers.

PREFERRED / BONUS EXPERIENCE
• Experience with real-time video streaming pipelines (NVENC/NVDEC, GStreamer, WebRTC) or causal streaming video diffusion (FIFO-Diffusion / CausVid-style rolling caches).
• Contributions to open-source GPU inference frameworks (`diffusers`, `xDiT`, `TensorRT`, `vLLM`, `SGLang`, or `FlashAttention`).

## Application Schema & Direct Apply (`apply_for_role`)

Candidates and authorized AI agents on behalf of a candidate can submit an application via the declarative WebMCP form tool `apply_for_role` on https://jobs.talent247.co/jobs/senior-inference-engineer-real-time-diffusion or via HTTP `POST https://jobs.talent247.co/api/public/jobs/senior-inference-engineer-real-time-diffusion/apply` (`application/json` or `multipart/form-data`). Non-browser HTTP agents must include header `X-Talent247-Intake-Token` obtained from `GET https://jobs.talent247.co/api/public/intake-token` (or automatically returned in `GET https://jobs.talent247.co/api/public/jobs/senior-inference-engineer-real-time-diffusion/form-schema`).

### Form Fields

- `full_name` (text, required): Full name
- `email` (email, required): Email
- `phone` (phone, optional): Phone
- `location` (text, optional): Location
- `linkedin_url` (url, optional): LinkedIn URL
- `cover_letter` (textarea, optional): Cover letter
- `consent` (checkbox, required): I consent to my data being processed for this application
- `cv_file` / `cv_base64` + `cv_filename`: Upload a CV in .pdf or .docx format (<= 10 MB) or provide an `https://` LinkedIn profile URL in `linkedin_url`.

### Alternative: Talent Pool Intake (`join_talent_pool`)

If this mandate is not an exact match, register once in the Talent247 Talent Pool at https://jobs.talent247.co/talent-pool (`https://jobs.talent247.co/talent-pool.md`) via WebMCP tool `join_talent_pool` or `POST https://jobs.talent247.co/api/public/talent-pool`.
