AI Engineer Lead

WHO WE ARE

DELVE Deeper is a global performance media agency where data, technology, and marketing intersect.

We help brands like UNICEF, Virgin Voyages, and Orange grow by using data, analytics, and automation to drive measurable results. Our teams work at the intersection of media, data science, and technology in a fast-paced, international environment.

ROLE OVERVIEW

The mandate is to take the Start Stop Scale (SSS) Agent from design through stable production across a multi-agent architecture, a data-heavy operating model, and a strict evaluation-first delivery process. The successful candidate will lead hands-on across LLM systems, data pipelines, connector infrastructure, context management, observability, and production quality.

The role is intentionally shaped for an AI engineering lead with a strong data engineering background. This person owns the data platform decisions that make the agent usable in production, including ETL orchestration, data contracts, vector storage, API reliability, and context formatting for LLM consumption.

WHAT THIS ROLE WILL BUILD

The immediate focus is the SSS Agent, a workflow system that combines automated weekly optimization reporting with ad hoc analysis via Claude projects, Skills, and MCP connectors. The build scope includes the core agent stack, human-in-the-loop approval flows, and the data layer that supports daily and weekly decisioning.

• A production-grade multi-agent system spanning planning, scale/stop, trendspotting, start, ad hoc analysis, report assembly, taxonomy verification, feedback verification, and daily callouts
• A data platform that supports 17 discrete SSS data components with clear ownership, freshness controls, and fit-for-purpose formatting for agent use
• A reusable MCP connector and tool layer for Semrush, SerpAPI, Slack, and media platform APIs
• A strict evaluation layer that defines success before build, measures quality during development, and monitors drift in production
• A feedback loop that captures trader approvals, rejections, rationale, and operational signals back into the system

KEY OUTCOMES AND DELIVERABLES

• Evaluation frameworks and success criteria defined before feature development starts
• All core SSS sub-agents shipped to stable production with clear input, output, and failure-mode documentation
• 17 data components designed, normalized, and governed through explicit data contracts
• Automated ETL or ELT pipelines supporting daily callouts and weekly optimization reporting
• Production-ready connector library with resilient retry logic, rate-limit handling, and fallback behavior
• Vector database or retrieval layer supporting long-term memory, similarity retrieval, and context assembly
• Monitoring, alerting, and quality regression checks across model outputs, connectors, and pipelines
• Slack-based human-in-the-loop workflows for approvals, feedback capture, taxonomy exceptions, and operational callouts

CORE RESPONSIBILITIES

1. Agent systems and platform architecture

• Design and implement the end-to-end architecture for the SSS Agent, with reusable patterns for orchestration, tool use, memory, and failure handling.
• Own context window strategy across the agent system, including chunking, retrieval, summarization boundaries, and contamination prevention.
• Build platform components that allow rapid iteration without compromising production standards, including internal libraries, shared services, and testing utilities.
• Translate product intent into technical architecture that preserves the integrity of the SSS decision logic.

2. Data engineering and information architecture

• Own the design of the data layer that powers agent decisions, including ingestion, normalization, storage, schema discipline, and data freshness.
• Build and maintain automated ETL or ELT pipelines pulling data from media platforms and internal sources on the cadence required by the product.
• Define and enforce data contracts between upstream systems and the agent layer so schema changes are managed deliberately, not discovered at runtime.
• Own vector storage, retrieval indexing, and context formatting so long-history performance data remains usable as volume grows.
• Design rate-limit controls, circuit breakers, retries, and fallback strategies so data gaps do not silently degrade agent quality.

3. Evaluation, experimentation, and quality control

• Define evaluation frameworks, success thresholds, and regression suites before any feature enters development.
• Run systematic prompt testing and AI-versus-human benchmarking to ensure agent outputs meet the quality bar required for live use.
• Establish confidence scoring, exception handling, and escalation logic for uncertain recommendations.
• Monitor production outputs for drift, quality regressions, hallucination patterns, and connector or data degradation.
• Set and enforce a clear production-readiness bar covering code quality, test coverage, documentation, and operational safeguards.

4. Delivery leadership and team direction

• Lead engineers as a hands-on technical lead: decompose work, review code, unblock execution, and maintain development velocity.
• Partner tightly with the Head of AI Transformation on backlog sequencing, scope realism, technical trade-offs, and build-vs-buy recommendations.
• Keep the build order aligned to dependency risk, prioritizing upstream components that unlock downstream reliability.
• Mentor engineers working in an AI-first codebase while maintaining a high standard for clarity, speed, and disciplined iteration.

5. Production operations and platform reliability

• Implement logging, tracing, monitoring, and alerting across model behavior, pipeline health, API usage, and user-facing failures.
• Respond to production issues quickly and drive root-cause fixes rather than surface-level patches.
• Set cost, rate, and usage limits deliberately so the system remains commercially viable as usage scales.

CANDIDATE PROFILE

Must-have experience

• 7+ years across software engineering, data engineering, ML engineering, or AI platform work, including direct ownership of production systems
• 2+ years leading technical delivery for complex systems, with evidence of setting standards and raising execution quality
• Strong Python and SQL, plus hands-on experience building APIs, services, and robust data pipelines
• Deep experience with ETL or ELT design, schema management, and data platform reliability in production
• Hands-on experience building or operating LLM applications, agentic systems, tool-calling workflows, or comparable AI application layers
• Experience with cloud infrastructure and containerized deployments (AWS, Azure, or GCP; Docker and ideally Kubernetes)
• Strong grounding in software engineering discipline, including testing, code review, CI or CD, observability, and incident response

Strongly preferred

• Experience with workflow orchestration and data tooling such as Airflow, Dagster, Prefect, dbt, Kafka, or similar platforms
• Experience with vector databases, retrieval systems, similarity search, and long-context data handling
• Familiarity with MCP, or equivalent integration layers that connect AI systems to enterprise tools and APIs
• Experience with performance marketing, ad-tech, or media platform APIs such as Google Ads, Meta, DV360, Semrush, or SerpAPI
• Experience shipping systems that mix model logic, deterministic business rules, and human approval flows
• Comfort translating messy business logic into precise technical rules without flattening the nuance

THIS ROLE IS NOT

• A workflow adoption or change-management role
• A low-code automation builder role
• A research-only AI scientist role
• A people manager removed from hands-on system ownership
• A generic prompt engineer role without responsibility for data, platform, and production quality

WHAT WE OFFER

• Work schedule: 12 pm- 8 pm
• A competitive salary and an annual bonus opportunities
• A promote from within culture and the chance to define your career growth
• Health and dental insurance (after trial period)
• 28 calendar PTO days
• Flexible sick days policy backed by full 100% short-time disability coverage
• Brand new office in Minsk, built and designed exclusively for DELVE
• English language tuition covered for specific positions
• Compensation of sports facilities (after trial period)
• Mental health reimbursement (after trial period)
• Generous employee referral bonuses
• Hybrid Working Model: Tuesdays, Wednesdays, Thursdays in office with the option to work from home on Mondays and Fridays