Architecting
Scalable Intelligence

Re-architected the Data Kit Deployment Orchestration Workflow, transitioning from a resource-intensive polling model to a scalable event-driven architecture. To enhance maintainability, I designed a task registry leveraging the Strategy Pattern, allowing for modular execution of post-deployment logic. This modernization drastically optimized system performance, reducing API overhead by 40x and accelerating notification delivery speeds by 97.5%.

Engineered a scalable clustering and labeling pipeline integrated directly into Data Cloud via Spark UDFs. This solution semantically groups complex contact reasons and generates descriptive labels for each cluster, exposing advanced logic through standard Data Transforms. This architecture directly bridged the gap between raw data and actionable product analytics.

Architected a robust evaluation framework to benchmark Small Language Models (SLMs) including Phi-4, Gemma, and Qwen-3b against a GPT-4 baseline for the Customer Signals Intelligence pipeline. I deployed high-performance inference endpoints using vLLM on Dockerized AWS EC2 (g6e.xlarge, Deep Learning Base OSS Nvidia Driver GPU AMI) instances. Leveraging an LLM-as-a-Judge (GPT-4.1) methodology, I rigorously measured critical trade-offs across performance scores, win+tie rates, consistency, and latency to validate model feasibility.

Engineered advanced ReAct and Chain-of-Thought (CoT) prompt templates to significantly reduce hallucinations while accurately extracting high-value signals like sentiment and contact reasons. Building on this foundation, I developed AgentForce Actions to drive automated case resolution and intelligent product description enhancement. To maximize business value, I extended platform extensibility, enabling customers to define and extract custom signals, persisting them as structured JSON within Data Cloud DMOs.

Played a pivotal role in architecting Customer Signals Intelligence (CSI) from the ground up. This platform addresses a critical business need: unifying fragmented customer interactions across Case, Voice, Chat, and other channels into a single, cohesive narrative. By harnessing the Generative AI capabilities of the AgentForce platform, we didn't just build a dashboard; we built a proactive engine that empowers agents to anticipate needs before they arise, setting a new standard for modern customer engagement.

Designed a scalable microservice architecture capable of handling 260M+ monthly events via Kafka. I optimized the system for high availability by leveraging PostgreSQL for reliable storage and Redis for low-latency caching, effectively removing I/O bottlenecks. The infrastructure was delivered via containerized deployments, supported by Grafana dashboards for real-time monitoring and issue resolution.