DeepSoft AI

Technologies

The stack behind every agent we ship

We build natively on the cloud and data platform you already run, and pick models, frameworks and tooling per use case, so your agents sit next to your data with your existing security, identity and governance.

Cloud & data platforms

Five platforms, one delivery team

DeepSoft AI at the centre of five platforms: AWS, Microsoft Azure, Google Cloud, Databricks and Snowflake.
PlatformAI & agent services we build withTypical work
AWS
Amazon BedrockBedrock AgentCoreSageMakerOpenSearch
Production agents, model hosting, fine-tuning and MLOps
Microsoft Azure
Microsoft FoundryAzure OpenAIAzure AI SearchAzure Databricks
Enterprise copilots, agents and RAG for Microsoft-first organisations
Google Cloud
Vertex AIGemini modelsGemini EnterpriseBigQuery
Gemini-powered agents, enterprise search and analytics assistants
Databricks
Agent BricksMosaic AIUnity CatalogMLflow
Agents on governed lakehouse data, evaluation and model serving
Snowflake
Cortex AICortex AgentsCortex AnalystSnowpark
AI and agents that run inside the data cloud, close to governed data

The agentic AI stack

Layer by layer, portable across clouds

Open, well-supported building blocks keep your agents portable and your options open.

01Foundation modelsChosen per task on quality, cost and latency
Anthropic ClaudeOpenAI GPTGoogle GeminiMeta LlamaMistralHugging Face models
02Agent frameworks & protocolsPlanning, tools, memory and multi-agent orchestration
LangGraphLangChainLlamaIndexModel Context Protocol (MCP)Agent2Agent (A2A)Bedrock AgentCoreFoundry Agent Service
03Retrieval & knowledgeRAG over documents, databases and APIs
PineconeFAISSpgvectorAzure AI SearchOpenSearchVertex AI SearchKnowledge graphs
04Evaluation & guardrailsQuality gates before and after release
RagasDeepEvalMLflow evaluationGENEVALPII & safety guardrailsHuman-in-the-loop review
05LLMOps & observabilityTracing, cost, drift and release management
MLflowLangSmithOpenTelemetrySageMaker PipelinesCI/CD for prompts & agents
06Data engineeringClean, governed data for agents to act on
Apache SparkDelta LakeUnity CatalogdbtAirflowKafkaSemantic layers
07Apps & infrastructureFrom API to user interface, deployed your way
PythonFastAPINode.jsReactDockerKubernetesTerraform
Models

Model-agnostic

We benchmark candidate models on your own data and switch when a better or cheaper option appears, without rewriting the agent.

Portability

Open tooling

LangGraph, LlamaIndex and MCP keep tools and agents portable, so a move between clouds is a deployment change, not a rebuild.

Quality

Measured delivery

Every release passes evaluation suites built with Ragas, DeepEval, MLflow and our own GENEVAL framework.

Choosing between platforms or models?

We compare options for your use case on quality, cost, latency and governance, then recommend one.