Project Vitals
- Role: Product Lead (Internal Platforms)
- Category: Enterprise Transformation & AI
- Strategy: Operational AI & RAG Implementation
- Context: S&P Global (SAM Platform)
1. The Strategic Thesis: The “Shadow Support” Bottleneck
The Secure Access Management (SAM) platform is the identity backbone for 85+ internal products, serving 1.8 million users and supporting a portfolio with a combined annual revenue of $2 Billion+.
However, as we scaled, our engineering team became trapped in “Shadow Support.”
- The Cost of Complexity: New product teams integrating with SAM found our documentation dense and complex.
- The Drain: The team was fielding 5-8 unsolicited Teams pings daily and losing 6-8 hours per week on “Consultation Calls.” This was untracked work that distracted senior engineers and support from shipping core features and working on other time sensitive support items.
- The Realization: We could not scale the platform to the next 50 products if we had to hand-hold every integration. We needed a way to free up our best engineers and support team members time by automating flows wherever possible.
2. The Solution: “Chat with Your Documentation”
I defined a vision to shift integration support from Manual Consultation to AI-Driven Self-Service. We built SAM Assist, a Generative AI chatbot embedded directly into our developer portal. It allows product teams to ask complex integration questions (e.g., “How do I implement OIDC for a my intelegence reports product?”) and get instant, code-ready answers based on our documentation.
3. The AI Architecture (RAG)
This wasn’t just a wrapper around ChatGPT. We built a robust Retrieval-Augmented Generation (RAG) system to ensure accuracy and data security. This was built on top of Spark Assist platform.
- Knowledge Base: We initially fed the model our technical PDFs.
- Automation: We later evolved the pipeline to integrate directly with Confluence. Now, whenever we update a documentation page, the system automatically re-indexes and updates the Vector Embeddings.
- The Result: The AI is always in sync with the latest API changes, requiring zero manual retraining.
4. Execution & Stakeholder Alignment
I partnered with the internal Spark Assist team to leverage their foundational AI models while my team provided the domain-specific data structure.
- The Pitch: I secured leadership buy-in by framing this not as a “cool AI project,” but as an Efficiency Lever. I demonstrated that “Shadow Support” was effectively costing us one Full-Time Employee (FTE) in lost productivity.
- The “Docs” Win: This also solved a user experience problem. New product teams no longer had to read 50-page wikis; they could simply query the bot to unblock themselves, significantly reducing their Time-to-Hello-World.
5. Outcome & Business Impact
- Zero-Headcount Scaling: We successfully onboarded new products to the platform without increasing our team size. The AI absorbed the complexity of the added volume.
- 40% Support Reduction: We saw a drastic drop in repetitive “How-to” tickets and DMs.
- Strategic Focus: By reclaiming the 6-8 hours/week previously lost to consultations, my engineering team shifted focus to high-value tasks, reducing turnaround time on critical platform upgrades.
6. Tech Stack
- Core: Python, RAG Architecture, Vector Database.
- Integration: Confluence API (Auto-Indexing).
- Platform: S&P Global Internal AI Cloud (Spark Assist).