Google Cloud & Gemini
We build agentic products and enterprise platforms on Google Cloud, combining Gemini capabilities, Gemini Enterprise integrations, Google ADK and speech technologies with Smriti's own application engineering.
A concept screen. A procurement lead asks which vendor contracts renew this quarter. The answer cites three approved sources. Three coordinated agents show their completed steps, and a proposed action — sending two price-review requests — waits for the procurement lead to approve, edit or send it back.
Where the work is
A capable model is the beginning of a product, not the product. Organisations need agents that work with their own knowledge, act in their own systems and involve a person when a decision needs one.
Our work is everything in between: the interface people use, the orchestration of specialised agents, the connections to approved knowledge and business systems, and the evaluation and deployment that make the whole dependable.
What we engineer on Google Cloud
Responsibility map
The product people use, and what makes it dependable
Google's technologies, which we build with
Gemini Enterprise
Role: Experiences and integrations
In our projects: Google's intranet search, AI assistant and agent platform for organisations. We connect it to company systems and build agents for it where a project uses it.
Keep in mind: Not a model-training or tuning service.
Gemini models
Role: Model capabilities
In our projects: Understanding, reasoning and generation, with the model chosen against the task.
Keep in mind: We name a specific model version only when a project actually uses it.
Gemini Enterprise Agent Platform Formerly Vertex AI
Role: Model access, tuning, agent hosting
In our projects: Where a project uses it: access to models, tuning for supported Gemini models, and Agent Runtime for hosting agents.
Keep in mind: Supervised tuning is available for specific Gemini models only — not every model, method or region.
Agent Development Kit (ADK)
Role: Agent development and orchestration
In our projects: The open-source framework we use to build, debug and deploy agents and the way they coordinate.
Keep in mind: Not Gemini-only: model-agnostic, and deployable to Agent Runtime, Cloud Run, GKE or other infrastructure.
Cloud Speech-to-Text and Cloud Text-to-Speech
Role: Voice interaction
In our projects: Speech recognition and synthesis where an application needs a voice interface.
Keep in mind: Voices, languages and models are confirmed for each application, not assumed.
Documents, records and business tools — integrated with permission
Model adaptation
Many tasks can be met with clearer instructions, the right context, retrieval from approved knowledge and a well-built harness. We evaluate each step against the task before considering the next.
When evaluation shows a model does need adapting, we check what the selected model supports. Supervised tuning is available for specific Gemini models only — some in preview or in limited regions — so it is confirmed per project, never assumed.
A clear task definition and the right context in every request.
Evaluate against the task — stop here if it is met
Answers grounded in the organisation's own sources rather than the model's general knowledge.
Evaluate against the task — stop here if it is met
Workflow rules, tool calls and checks around the model.
Evaluate against the task — stop here if it is met
Supervised tuning, only when evaluation shows the earlier steps fall short and the selected Gemini model supports it.
What we build
Capability overview
Agents that work with an organisation's own knowledge and software, built on Google Cloud with Gemini models, Gemini Enterprise and the Agent Development Kit.
A capability overview describes what we engineer in this kind of project. It is not a case study: it names no customers and claims no results.
Google Cloud & Gemini