RobinRivera

Sales teams burn hours manually researching and qualifying leads and writing personalized outreach - work that simply doesn't scale.

Built a multi-agent pipeline (research -> score -> copywrite) orchestrated by a FastAPI worker running on an NVIDIA Jetson at the edge, securely proxied from a Next.js app. Agent prompts and JSON schemas are versioned for reproducible behavior, and campaigns soft-fail gracefully while self-healing stalled jobs.

• →End-to-end automation from ICP definition to qualified leads with ready-to-send outreach copy.
• →Resilient architecture: campaigns survive worker downtime and auto-recover zombie jobs.
• →Per-lead scoring plus generated cold-email and cold-call copy tailored to each prospect.

Real estate investors lack a fast, rigorous way to compare opportunities against the true opportunity cost of capital, and running AI-driven analysis on every interaction is too slow and expensive.

Designed a brain/muscle architecture: AI runs once offline to produce vetted financial assumptions, while a deterministic, pure finance engine recomputes every indicator instantly on the client - eliminating runtime AI cost and latency. Live CETES rates are pulled from Banxico's SIE API, and 7-page institutional PDF reports are generated server-side on demand.

• →What-if calculator runs fully client-side with zero per-calculation AI cost or latency.
• →Every analysis is benchmarked against a live risk-free rate (CETES via Banxico).
• →On-demand institutional PDF reports and invite-only access control with role-based permissions.

Knowledge and follow-ups scattered across chats and notes never get acted on, and generic assistants don't know your context or sound like you.

Wired a Claude-powered agent into an Obsidian vault with voice dictation and WhatsApp delivery through local agents (launchd), plus a companion edge AI agent running on an NVIDIA Jetson for shared team briefs. A voice profile built from real message samples lets it draft in my own register.

• →Daily briefs and voice-triggered capture and recall delivered straight to WhatsApp.
• →Runs locally for privacy - only keyword-triggered dictations are ever processed.
• →The assistant writes and recalls in my own voice, not a generic tone.

Local automotive repair shops struggled with manual, error-prone processes for managing operations, customer records, service history, and billing.

Built a full-stack platform leveraging AI to streamline shop operations, automating workflows for scheduling, invoicing, inventory, and customer management.

• →Successfully deployed in 5 automotive repair shops.
• →Reduced operational overhead and billing errors significantly across all deployments.
• →Enabled shop owners to focus on service quality instead of administrative tasks.

People struggle to consistently track their nutrition due to complex apps and manual data entry, making it hard to meet dietary and health goals.

Created an intelligent application powered by OpenAI that automates calorie and macronutrient tracking through natural language and image recognition, making nutrition logging effortless.

• →Improved user experience through end-to-end process automation.
• →Drastically reduced the time needed to log meals compared to traditional apps.
• →Helped users build consistent dietary habits with a frictionless tracking experience.