Google ADK Agents on Gemini: Integration and Deployment

Problem: Multi-agent systems as a puzzle of 15 libraries

You assembled a team of three ML engineers, spent a month integrating LangChain, AutoGen, and your own orchestrator. The result — a mess of callback functions, p99 latency over 3 seconds, and not a single agent in production. We see this all the time.

Google ADK is a framework that solves this pain at the architecture level. It doesn't require gluing five libraries together: agent hierarchy (LlmAgent, SequentialAgent, ParallelAgent) comes out of the box, and deployment to Vertex AI is one command. In this article, we'll show how we use ADK in real projects and how it saves up to 70% development time and significantly reduces API costs.

What problems does Google ADK solve?

Chaotic agent coordination

Note: when you have more than two agents, managing their interaction becomes hell. Without a standard pattern, each agent calls another via API, breaking the chain when one link fails. ADK offers three clear patterns: sequential pipeline (SequentialAgent), parallel execution (ParallelAgent), and hierarchical orchestrator. This covers 90% of business scenarios without custom infrastructure.

Scattered context and memory loss

Many frameworks don't clean the context window — agents clutter the dialogue history, increasing token consumption by 40-50%. ADK automatically manages sessions through caching and cleans unnecessary data. In a project for FMCG, this reduced Gemini API costs by 35%, saving the client about $8,000 per month.

Long path from prototype to production

Moving an agent from Jupyter Notebook to production typically takes weeks — you need to write API wrappers, set up monitoring, handle CORS/authentication. ADK natively deploys on Vertex AI Agent Builder: one command adk deploy — and the agent is available as a REST endpoint. Official ADK documentation describes it in 3 steps.

How we do it: stack and case

Stack

• Gemini 2.0 Flash (low-latency), Gemini 2.0 Pro (complex reasoning)
• Google ADK v0.9 (latest stable version)
• Google Search Grounding, Vertex AI Search, custom FunctionTool
• Vertex AI Agent Builder, GKE for high-load scenarios

Real-world case: competitor monitoring system for FMCG

The client is a large food manufacturer. Their marketing analytics department spent 3 man-days per week tracking 15 competitors: prices, news, new products. We built a multi-agent system on ADK in 10 working days.

Architecture:

• ParallelAgent launches 15 sub-agents in parallel — each monitors one competitor.
• SequentialAgent passes data to a trend aggregator, then to a report generator.
• Output: weekly digest for the CMO in executive summary format.

Results:

Analysts shifted to strategic tasks. Savings were $10,000 per month on API alone. In another project for a retailer, we reduced token costs from $12,000 to $3,000 per month.

How to create a basic agent on Google ADK: step-by-step guide

1. Install Google ADK and set up environment: pip install google-adk.
2. Create a function tool, e.g., for fetching a stock price.
3. Define LlmAgent with Gemini model and tool list.
4. Run the agent via Runner and test in a session.

from google.adk.agents import LlmAgent
from google.adk.tools import google_search, FunctionTool
from google.adk.runners import Runner
from google.adk.sessions import InMemorySessionService

def get_stock_price(ticker: str) -> dict:
    """Get current stock price.

    Args:
        ticker: Stock ticker (e.g., GOOGL, AAPL)

    Returns:
        dict with price, change, and volume
    """
    data = finance_api.get_quote(ticker)
    return {
        "ticker": ticker,
        "price": data["price"],
        "change_percent": data["change_percent"],
        "volume": data["volume"],
    }

stock_tool = FunctionTool(func=get_stock_price)

research_agent = LlmAgent(
    name="market_researcher",
    model="gemini-2.0-flash",
    instruction="""You are a financial market analyst.
Research market data and provide structured analysis.
Always use tools to get current data.""",
    tools=[google_search, stock_tool],
    output_key="research_result",
)

session_service = InMemorySessionService()
runner = Runner(
    agent=research_agent,
    app_name="financial_analysis",
    session_service=session_service,
)

How does Google ADK compare to alternatives?

ADK wins in scenarios where deployment speed and out-of-the-box solutions matter. For custom low-level tasks, LangChain is better — but we rarely see such tasks in commercial projects.

# Example: orchestrator with two sub-agents
from google.adk.agents import LlmAgent, SequentialAgent

analyst = LlmAgent(name="analyst", model="gemini-2.0-flash", instruction="Analyze data")
writer = LlmAgent(name="writer", model="gemini-2.0-pro", instruction="Write report")

orchestrator = SequentialAgent(
    name="orchestrator",
    agents=[analyst, writer],
    output_key="final_report"
)

Our process

• Analysis (2-5 days): Understand business processes, identify agent application points. Gather requirements for latency, traffic, integrations.
• Design (1-3 days): Design agent hierarchy: orchestrator, sub-agents, data schema.
• Implementation (3-10 days): Write code, connect tools, configure grounding.
• Testing (2-3 days): Load tests (p99 latency, throughput), error resilience checks.
• Deployment (2-3 days): Deploy on Vertex AI, set up CI/CD, monitoring (Cloud Monitoring, logs).
• Handover (1-2 days): Documentation, team training, SLA.

What's included in the work

We audit current processes to identify bottlenecks, design the agent system architecture, implement code using Google ADK, Gemini API, and FunctionTool, then deploy to Vertex AI or GKE with CI/CD. You receive documentation, your ML engineer training, and two weeks of post-production bug fixing and optimization.

Estimated timelines

Cost is calculated individually for your scenario — contact us for a project estimate within 1-2 days.

Why choose us?

• 5+ years of AI/ML experience, 30+ completed agent integration projects.
• Certified Google Cloud specialists (Vertex AI, Gemini certifications).
• Result guarantee: we fix KPIs in the contract (latency, accuracy, token costs).
• Open communication: you get repository access and see progress in real-time.

Ready to discuss your task? Get a consultation on ADK implementation — we'll choose the optimal configuration and show a prototype in 2-3 days. If you want to dive deeper yourself, read the official Google ADK repository on GitHub — many examples there.

LLM Development: Fine-Tuning, RAG, Agents, and Production Deployment

Using GPT‑4 or Claude 3.5 Sonnet through a public API is not a solution — it's just a tool. When the requirement is to "make it like ChatGPT, but on our data," there is a real engineering challenge behind it: from prompt engineering to training a 70B model on your own infrastructure. End-to-end LLM solution development is a complex stack, and we have been doing it for over 5 years. During this time, we have completed over 20 projects in generative AI: from RAG systems for legal departments to custom support agents. Where exactly your task falls depends on data, latency requirements, budget, and how critical confidentiality is.

A typical situation: the client has already tried ChatGPT, but results are unstable — sometimes accurate, sometimes hallucinating. Or they need integration into a corporate portal while complying with security policies. Let's break down each layer of the stack in detail — from RAG to production deployment.

Why Do RAG Systems Break and How to Fix It?

RAG (Retrieval-Augmented Generation) looks simple: find relevant documents, put them in context, get an answer. In practice, it fails in several places.

Chunking without overlap. Classic mistake: chunk_size=512, overlap=0. If the answer lies across two chunks, retrieval won't find either with sufficient confidence. Solution: overlap 15–25% of chunk_size, or better yet, sentence-aware splitting with spaCy or NLTK instead of naive character splitting.

Poor embedder. text-embedding-ada-002 is good for general use, but on legal or medical texts, specialized models like E5-large-v2, BGE-M3, or fine-tuned sentence-transformers on domain data outperform it. Recall@5 differences can be 15–25%.

No re-ranking. Vector search optimizes for speed, not relevance. A cross-encoder re-ranker (ms-marco-MiniLM-L-6-v2, bge-reranker-large) after initial retrieval improves top-3 accuracy with acceptable latency (+50–150ms). This is often more impactful than improving the embedding model.

Hybrid search. Dense vectors alone work poorly on exact queries: names, SKUs, codes. BM25 (sparse) finds exact matches but misses semantics. Hybrid via RRF (Reciprocal Rank Fusion) is the optimal compromise. Qdrant, Weaviate, and pgvector 0.7+ support hybrid search natively.

When to Fine-Tune Instead of Prompt Engineering?

Prompt engineering solves ~70% of LLM adaptation tasks for a domain. The remaining 30% require fine-tuning. Three indicators: the model ignores a specific output format even with detailed prompting; the task requires deep knowledge of specialized vocabulary (medicine, law); you need to significantly reduce token costs by replacing a large model with a smaller specialized one.

LoRA and QLoRA are the standard for SFT. LoRA adds trainable low-rank matrices to attention layers. A typical configuration for Llama-3 8B: r=64, lora_alpha=128, target_modules=["q_proj","v_proj","k_proj","o_proj"] yields ~0.8% trainable parameters, training on one A100 40GB. QLoRA adds 4-bit quantization (NF4) and allows fine-tuning 70B models on two A100 40GB, though speed drops by half compared to bf16.

DPO instead of RLHF. Direct Preference Optimization requires only (chosen, rejected) pairs, not scalar reward signals. DPOTrainer from the trl library (Hugging Face) implements it in a few dozen lines.

Common mistake. A dataset of 500 examples, 5 epochs, validation loss 0.8 — seems fine. But on test, the model degrades on general instructions. Cause: catastrophic forgetting. Solution: add 10–20% general instruction-following examples (Alpaca, FLAN) to the training set to preserve original capabilities.

How to Choose a Base Model: 8B or 70B?

For most tasks, fine-tuning an 8B model is sufficient. 70B is needed when deep reasoning is required or the 8B baseline does not reach the required quality even after fine-tuning. Inference cost for Llama-3 8B via vLLM on A100 is efficient; the exact cost depends on volume.

What Does PagedAttention Bring to Production?

vLLM is the first choice for serving open-source models. PagedAttention is the key technical innovation: KV-cache is managed like virtual memory in an OS, without fragmentation. This yields 2–4x higher throughput compared to naive HuggingFace Transformers inference. The vLLM documentation confirms that continuous batching and PagedAttention are the standard for high-load LLM services.

Typical numbers on A100 80GB for Llama-3 8B (bf16): 400–600 req/s, P50 latency 200–400ms, P99 latency 600–900ms at concurrency 64. For 70B on two A100 with tensor parallelism: 80–120 req/s, P99 latency 1.5–2.5s. AWQ or GPTQ quantization reduces memory consumption by 2x with quality loss within 1–3%.

Multi-Agent Systems

Agents are LLMs with access to tools: search, code execution, API calls, database interaction. Common patterns:

• ReAct (Reason + Act): the model reasons → chooses a tool → observes the result → reasons again. LangChain and LlamaIndex implement it out of the box.
• Multi-agent orchestration: multiple specialized agents with a coordinator on top. Example: coordinator → researcher (search + summarization) → coder (code generation and execution) → critic (verification). Tools: AutoGen (Microsoft), CrewAI, custom implementation on LangGraph.

In production, agent systems are non-deterministic. Essential: guardrails, step limits, logging of each step, human-in-the-loop for critical actions.

How We Work: Stages, Timeline, Deliverables

What Is Included

We deliver:

• Technical documentation (model card, configs, deployment instructions)
• Access to infrastructure (code repository, trained weights)
• 1 month of post-deployment support (consultations, bug fixes)
• Customer team training (2–3 sessions on system operation)

Timeline: basic RAG prototype — 1–2 weeks. Fine-tuning with customer data — 3–6 weeks (including data preparation). Production system with monitoring and retraining — 2–4 months. Cost is calculated individually based on data volume, model complexity, and infrastructure requirements.

We guarantee the quality of the final model with performance benchmarks and ongoing monitoring. Our engineers have hands‑on experience with dozens of production LLM systems.

Want to evaluate your project? Leave a request — we will prepare a preliminary summary within 1–2 business days. Or get a consultation on choosing the approach: RAG, fine-tuning, or hybrid — we will tell you what works best for you. Contact us to discuss your LLM development needs. Schedule a free consultation today.