Senior AI Platform Engineer

You will own the AI layer that powers Synagi's agents-from vector databases and retrieval-augmented generation (RAG) pipelines to fine-tuning compact transformer models and classic ML solutions where they make sense. Your work will turn raw data into fast, reliable intelligence that scales with our product ambitions. Core Responsibilities

• Vector databases - Design schemas, sharding strategies, and ANN indexes (Milvus, Vespa, or pgvector) to store and query billions of embeddings.
• RAG pipelines - Build and maintain end-to-end retrieval workflows: query rewriting, hybrid BM25 + vector search, and re-ranking for fact-grounded answers.
• Model creation & fine-tuning - Train or adapt lightweight transformer models using techniques such as LoRA; develop classic ML models when they outperform deep nets.
• MLOps - Containerise AI workloads with Docker, deploy and scale them on Kubernetes, and automate training/evaluation workflows.

• 3+ years of production machine learning experience in Python; shipped at least one transformer-based model.
• Proficient with PyTorch or JAX for custom model development and fine-tuning.
• Practical experience with vector databases and RAG techniques.
• Comfortable with CUDA tooling for debugging and optimising GPU workloads.
• Able to design and train ML models from scratch for small-parameter or classical ML tasks.

Nice-to-Haves

• Experience with DeepSpeed or vLLM for efficient inference serving.
• Familiarity with LangChain or LlamaIndex for rapid agent prototyping.
• Interest in decentralised or edge deployments (eg, WASM at the edge) for ultra-low-latency inference.

We offer highly competitive salary, early-stage equity, and an opportunity to be the backbone of synergetic general intelligence.

At Synagi, we are pushing the frontier of distributed and decentralised AI agents. Our research spans vector-driven retrieval systems, agentic swarms, and resource-efficient multi-agent architectures-all with a sharp focus on real-world performance and human-in-the-loop alignment. We explore scalable, context-aware multi-agent designs that outperform monolithic approaches and keep compute costs in check.