Solutions Architect (AI)
Summary
Solutions Architect at Liquid Intelligent Technologies (Cassava AI) designing and delivering enterprise AI infrastructure — accelerated compute, model serving, inference/RAG design, Kubernetes orchestration — from proof of value to production, plus pre-sales technical content for bids and tenders. Core stack: NVIDIA AI stack (CUDA, Triton, NIM, NeMo), Kubernetes, Python, Terraform.
ROLE PURPOSE
Deliver solution architecture for enterprise AI workloads across Cassava AI’s three delivery streams: AI Factories, AI Infusion and AI Diffusion. Translate business requirements into deployable reference architectures and production-ready implementations spanning accelerated compute, inference and retrieval design, model serving, workload orchestration, observability, security controls and lifecycle management, integrated into existing enterprise data, identity, network and application estates.
Technical solutioning behind tenders and carry workloads from proof of value through to production, including platform provisioning, deployment automation, validation, performance tuning and operational readiness.
ROLE DESCRIPTION
Solution Architecture and Design - Produce reference architectures and solution designs that a build team can execute from.
Solution Delivery and Implementation - Build, deploy and optimise AI workloads on the platform, and integrate them into customer environments to agreed specification.
Proof of concept to production conversion, deployment to specification and schedule, model and inference performance at handover, customer environment integration, production readiness (runbook, monitoring and rollback documented before go-live)
Pre-Sales, Bids & Client Engagement
Supply the technical content and technical credibility that convert qualified opportunities, across discovery, proposals, bids and tenders.
Enablement, Standards and Knowledge Transfer
Raise the technical standard of the team and the reuse rate of its work through patterns, review and training. NVIDIA and AI/ML stack - maintain certifications and report material stack changes to the team.
Key Decisions & Accountabilities
Reference architecture and technology selection for an assigned solution, within approved platform guardrails
Model, framework and serving stack for a given workload
Compute, storage and network sizing carried into solution designs and pricing inputs
Technical acceptance criteria and production readiness at handover
Accountable for the technical integrity and buildability of every solution design issued under their name, sizing assumptions and the stated performance basis in client-facing architecture, compliance with data sovereignty, data residency and platform guardrails in all designs, quality of technical content in assigned bids, tenders and proposals, technical handover from proof of concept to production, including documentation and knowledge transfer
ROLE REQUIREMENTS
B. Degree in Computer Science, Engineering, Data Science or a related field mandatory
M. Degree in Computer Science, Machine Learning, Data Science or a related field advantageous
NVIDIA certification or related certification: NCA or NCP AI Infrastructure, or equivalent
Kubernetes or cloud architecture certification: CKA, CKAD or hyperscaler architect level
4 years' AI and machine learning solution design and delivery in production environments with two years' hands on GPU infrastructure for model training or inference serving
Production-grade Python software development
Client-facing solution architecture, including requirements workshops and technical presentation
MLOps or platform engineering: Kubernetes, containerised model deployment, CI/CD for models
NVIDIA AI software stack: NVIDIA AI Enterprise, NIM microservices, NeMo, Triton Inference Server, CUDA, TensorRT-LLM
GPU compute architectures across providers
Generative AI patterns: retrieval augmented generation, fine-tuning, quantization, inference optimisation, guardrails and evaluation
Classical machine learning and deep learning: PyTorch, TensorFlow, scikit-learn, model selection and evaluation
Containerization and orchestration: Docker, Kubernetes, GPU Operator, scheduling and multi-tenancy
Data architecture for AI: ingestion pipelines, feature engineering, vector databases, parallel file systems and object storage
Workload sizing and total cost of ownership modelling for GPU infrastructure
Security and compliance for AI platforms: tenant isolation, data sovereignty and residency, POPIA
Specialisation in an AI domain such as MLOps, LLMOps, Orchestration, GenAI
AI/ML Workload sizing, benchmarking and performance tuning. AI/ML Requirements elicitation and translation into technical architecture
Infrastructure as code and automation: Terraform, Ansible, Helm and secondary programming language (Pyspark, Go, Java or C++) advantageous
Skills
- Acceptance Criteria
- AI
- Ansible
- Automation
- CI/CD
- Cloud
- Containerization
- C++
- CUDA
- Data Science
- Deep Learning
- Docker
- Feature Engineering
- Fine Tuning
- Generative AI
- GPU
- Helm
- Infrastructure as Code
- Java
- Kubernetes
- LLM
- LLMOps
- Machine Learning
- Microservices
- MLOps
- Model Deployment
- Nim
- Observability
- Pre-sales
- Proof of Concept
- PySpark
- Python
- PyTorch
- Quantization
- RAG
- Requirements Elicitation
- scikit-learn
- Solution Design
- TensorFlow
- TensorRT
- Terraform
- Triton
- Vector Databases
Sign in to read and join the discussion.