Lead AI Researcher & Architect (Inference Infrastructure & Systems)About Siellium Global Labs
Siellium Global Labs is a deep-tech startup engineering a zero-retention infrastructure layer across sovereign enterprise networks. We are building a local-first execution layer powered by a proprietary, dynamic Mixture-of-Agents (MoA) orchestration framework.
The Role
We are seeking a visionary and deeply technical Lead AI Researcher & Architect to design, build, and scale our next-generation local inference engines. This is a foundational, heavy-infrastructure leadership role. You will own the architectural blueprint and mathematical algorithms that directly manipulate GPU memory, maximize token throughput, and drive our core Mixture-of-Agents (MoA) execution layer.
As a player-coach, you will split your focus between pioneering novel optimization algorithms and leading an elite team of AI inference engineers. Your mission is to push the absolute limits of local hardware, architecting systems that dynamically hot-swap domain-specific expert models natively within VRAM at sub-millisecond latencies.
Core Responsibilities
- Architecture & System Design: Design the end-to-end blueprint for our high-concurrency local inference pipeline. Architect the integration between VRAM-optimized execution layers and our high-throughput Python/FastAPI routing gateways.
- Algorithmic Innovation: Develop and implement novel mathematical compression algorithms, structured matrix pruning techniques, and advanced quantization strategies (e.g., custom AWQ/GPTQ kernels) to squeeze frontier-class model logic into constrained enterprise hardware budgets.
- Dynamic MoA Orchestration: Create the architectural framework for asynchronous, zero-latency hot-swapping of specialized Low-Rank Adapters (LoRA) over pinned foundational base models (e.g., Llama-3-8B).
- Technical Team Leadership: Hire, mentor, and lead an elite team of AI inference and backend infrastructure engineers. Establish rigorous engineering standards, code reviews, and research methodologies.
- Throughput & Concurrency Optimization: Drive the research and implementation of advanced continuous batching parameters, custom PagedAttention layers, and non-blocking event loops to maximize hardware TFLOPS and eliminate memory bottlenecks.
Key Requirements & Technical Stack
- Pioneering Research & Architecture: Proven track record of designing and deploying production-grade AI inference architectures or publishing peer-reviewed research in model compression, quantization, or distributed inference.
- Leadership Experience: 3+ years of experience managing, mentoring, or technically leading high-performing teams of AI researchers or infrastructure engineers in a fast-paced environment.
- Deep Learning Systems: Expert-level mastery of PyTorch, the Hugging Face ecosystem (Transformers, PEFT), and low-level CUDA mechanics.
- Advanced Inference Engine Optimization: Deep architectural understanding of vLLM, PagedAttention, continuous batching, and speculative decoding mechanics.
- Hardware & Quantization Expertise: Direct experience developing or heavily modifying quantization frameworks (AWQ, GPTQ, GGUF) and optimizing model footprints for enterprise local hardware or cloud GPU instances (Vast.ai, RunPod, AWS).
- Languages & Concurrency: Mastery of production-grade Python (3.11+) and advanced asynchronous programming (asyncio). Knowledge of low-level languages (C++/CUDA) for custom kernel development is a massive plus.
Preferred Qualifications
- Degree in Computer Science, Electrical Engineering, Mathematics, or a related field with a focus on Deep Learning systems.
- Familiarity with state-of-the-art structured pruning libraries or research implementations (e.g., SparseML, Wanda).
- Experience architecting highly concurrent, production-ready web gateway frameworks (FastAPI) acting as routing layers for distributed LLM backends.
Pay: ₹2,000.00 - ₹5,000.00 per hour
Benefits:
Work Location: Remote
