← Back

Business Intelligence Engineer II · AWS Marketplace

From scheduled narratives to self-service agentic systems

LLM-powered narrative automation → Agentic systems in production

The Evolution

I started at AWS building data products that earned me a promotion to L5. Then I shifted to AI: first with narrative automation that generated insights on a schedule, now with agentic systems that answer questions on-demand. Same data infrastructure, increasingly intelligent layer on top.

Narrative Automation (Shipped)

Problem: Business teams spent significant time manually aggregating data from multiple dashboards to produce monthly performance narratives.

Solution: I pioneered the first LLM-powered narrative automation for AWS Marketplace reporting. Engineered 20+ aggregated data tables, developed context-rich prompts, and built automated pipelines that generated business narratives directly from structured data. Delivered monthly via email to stakeholders.

Impact: Eliminated manual aggregation bottleneck. Framework adopted by other team members for additional reporting workstreams.

Limitation:Narratives were scheduled and one-way. Teams couldn't ask follow-up questions or get insights on-demand.

Agentic Systems (Shipping Now)

Agents are the next evolution: self-service, proactive, conversational. Same data foundation, smarter interface.

Multi-dimensional renewals agent (in testing)

Trained on the same data tables as the narrative system, but with more sophisticated prompts designed for on-demand answers. Answers questions not obvious from dashboards: "Why did this metric move?" "What's driving the trend in seller X?" "Which customers are at risk?"

Key difference: Self-service instead of scheduled. Users ask when they need answers, not when the report is generated. Monitors metrics across customer, seller, and subscription views continuously.

Oncall support agent (in production)

Built for my BI team based on 3+ years of oncall experience. Handles data access requests, metrics definitions, and data discovery autonomously.

The problem: As an oncall engineer, I answered the same questions every week: "How do I access this data?" "What does this metric mean?" Stakeholders waited hours for repetitive answers.

Solution: Created knowledge bases from institutional knowledge, integrated live data tables, and designed prompts that route queries to the right capability. Cut response time from hours to seconds.

Why this matters: Proactive problem-solving. I identified the bottleneck, built the solution, and shipped it without being asked.

The Through-Line

Data products → Narrative automation → Agentic systems.

  • Data products: Built dashboards and decision systems (5,000+ users, 40 hrs/week saved). Earned L5 promotion.
  • Narrative automation: Added AI layer on top. Automated "what happened" insights on a schedule.
  • Agentic systems: Made AI interactive and self-service. Answers "why" and "what about X" on-demand.

The data tables I built for dashboards became the foundation for narratives. The prompts I developed for narratives became the basis for agents. Each layer compounds.

Product Decisions

Built reusable infrastructure

Data tables and prompt frameworks scaled from narratives to agents without rework. Higher upfront complexity, but every subsequent system was faster to build.

Grounded AI in structured data

Every system starts with validated, structured data. Prompts guide reasoning, but the foundation is deterministic. This builds trust and ensures accuracy.

Solved problems proactively

The oncall agent wasn't requested. I saw the problem and shipped the solution. Same with narrative automation: identified the bottleneck before it became a formal requirement.

Moved from scheduled to self-service

Narratives were limited by their cadence. Agents eliminate that constraint: users get answers when they need them, not when the report is scheduled.

Impact

  • First LLM-powered narrative automation for AWS Marketplace reporting
  • Framework adopted org-wide for additional reporting workstreams
  • Multi-dimensional renewals agent in testing: self-service insights across customer, seller, and subscription views
  • Oncall support agent in production: hours to seconds for data access and discovery
  • Built on data infrastructure powering 5,000+ users