Job Description:
LLM Engineer — Agentic Solutions
Marketing Intelligence Platform · Google Cloud · Pune / Remote · Full-time
Experience : 5 – 9 years
Location : Pune / Hybrid / Remote
Employment : Full-time
Seniority : DCF35 or DCF40
We are building a next-generation Marketing Intelligence Platform that turns Meta Ads campaign data into autonomous, actionable intelligence. At the core of this platform is a multi-agent system built on Google Cloud's Agent Development Kit (ADK) — capable of answering complex marketing questions in natural language, advising on campaign configuration, and autonomously detecting and alerting on performance anomalies.
As a Senior LLM Engineer specialising in Agentic Solutions, you will design, build, and operate the agent fleet that powers this platform. You will work at the intersection of large language model engineering, GCP data infrastructure, and production software delivery — owning both the agent logic and the underlying tool integrations that give agents real capability.
This is a high-ownership, high-visibility role. You will be the primary technical decision-maker for agent architecture and will collaborate directly with data architects, ML engineers, and marketing stakeholders to translate business requirements into production-grade agentic systems.
-
Design and implement a multi-agent hierarchy using GCP Agent Development Kit (ADK) — LlmAgent, SequentialAgent, ParallelAgent, LoopAgent — selecting the right class for each control flow pattern
-
Write and tune system prompts, tool schemas, and output schemas that give agents reliable, grounded behaviour across hundreds of distinct question types
-
Build the tool layer (Python functions) that connects agents to BigQuery, Vertex AI Search, Vertex AI prediction endpoints, Meta Marketing API, Pub/Sub, and Secret Manager
-
Implement multi-mode orchestration: chatbot mode, advisor mode, and comparator mode from a single OrchestratorAgent entry point
-
Design session memory and conversation context management using ADK session services, backed by Cloud Firestore for persistent cross-session history
-
Own the prompt engineering lifecycle: write, version-control, A/B test, and iterate on system prompts for all agents
-
Design and maintain a golden evaluation set (100+ Q&A pairs) and an LLM-as-judge evaluation pipeline that blocks deployments below quality thresholds
-
Implement ReAct (Reason + Act) loop patterns, self-correction via critic agents, and reflection loops to improve answer quality on multi-step queries
-
Establish guardrail layers — input classification, output validation, PII scrubbing — appropriate for a marketing data context
-
Deploy conversational agents to Vertex AI Agent Engine; deploy infrastructure agents as Cloud Run Jobs triggered by Cloud Scheduler
Build and maintain CI/CD pipelines using Cloud Build: eval gate container build staging deploy- production promotion
-
Configure IAM service accounts with minimum required permissions across BigQuery, Vertex AI, Discovery Engine, Pub/Sub, and Secret Manager
-
Instrument agents with structured Cloud Logging traces and set up Cloud Monitoring dashboards and alert policies for latency, error rate, and eval pass rate
-
Own the BigQuery tool layer — write and optimise SQL queries that back agent tool calls, manage gold dataset schemas, and ensure query cost efficiency
-
Maintain and grow the Vertex AI Search knowledge base (Meta best-practice docs, internal campaign guides, KB snippets)
-
Collaborate with the ML team to wire Vertex AI prediction endpoints (ROAS lift, CVR-by-bid, spend optimiser) into the InsightAgent tool registry
-
Design and populate the Config Store tables in BigQuery (best-practice configs, scoring rules) used by the ComparatorAgent
-
Define agent engineering standards: code structure, tool schema conventions, prompt versioning, eval thresholds, and deployment gates
-
Conduct code and prompt reviews; provide technical mentorship to mid-level engineers joining the agentic workstream
-
Translate ambiguous marketing requirements into well-scoped agent capabilities — writing technical specs that stakeholders can validate before implementation
-
Proactively identify capability gaps (e.g. missing ML model, incomplete KB, schema drift) and drive resolution across engineering teams
-
5+ years of software engineering experience, with at least 2 years in production LLM or AI systems
-
Demonstrable experience building multi-agent systems: agent hierarchies, tool use, ReAct loops, self-correction patterns — in production, not just prototypes
-
Strong prompt engineering skills: system prompt design, few-shot example construction, output schema enforcement, and structured output parsing
-
Hands-on experience with at least one agentic framework — GCP ADK, LangChain, LangGraph, CrewAI, AutoGen, or equivalent
-
Understanding of LLM evaluation: golden datasets, LLM-as-judge, RAGAS or similar metrics, A/B testing prompts against quality thresholds
-
Production experience with Vertex AI — model endpoints, Vertex AI Search (Discovery Engine), and ideally Vertex AI Agent Engine or Reasoning Engine
-
Strong BigQuery skills: complex SQL, partitioning and clustering strategy, query cost optimisation, schema design for analytics workloads
-
Experience with Cloud Run, Cloud Functions, Cloud Scheduler, Pub/Sub, and Cloud Build for production workloads
-
Familiarity with GCP IAM, Secret Manager, and Cloud Logging / Cloud Monitoring for secure, observable deployments
-
Python proficiency: async programming, Pydantic models, FastAPI or similar, clean modular code structure
-
Strong understanding of REST APIs, JSON schema design, and API integration patterns (webhooks, polling, retry logic)
-
Experience with Docker, container registries (GCR / Artifact Registry), and CI/CD pipelines
-
Git-based development workflow: branching strategy, PR reviews, automated testing in CI
-
Direct experience with GCP Agent Development Kit (ADK) — LlmAgent, SequentialAgent, ParallelAgent, LoopAgent, agent_tool, FunctionTool
-
Experience with the Meta Marketing API — campaign management, insights endpoints, webhook/HMAC integration, rate limit handling
-
Background in adtech, marketing analytics, or performance marketing data — familiarity with ROAS, CPM, CVR, incrementality measurement concepts
-
Experience building RAG (Retrieval-Augmented Generation) pipelines: chunking, embedding, vector store indexing, hybrid retrieval, reranking
-
Knowledge of ML model serving on Vertex AI: online prediction endpoints, model versioning, A/B traffic splitting
-
Google Cloud Professional certifications: Cloud Architect, ML Engineer, or Data Engineer
-
Experience with Gemini API / Gemini 1.5 Pro / Flash model families, including function calling and structured output schemas
-
Prior experience in a marketing intelligence, business intelligence, or data product engineering context
Location:
DGS India - Mumbai - Thane Ashar IT Park
Brand:
Merkle
Time Type:
Full time
Contract Type:
Permanent
