Briefings over dashboards
The system delivers a prioritized narrative at the right moment instead of creating another interface that demands attention.
Selected work
Case study · Executive AI agent · In development
JARVIS
A context-aware executive assistant designed to turn goals, commitments, and connected information into clear priorities and coordinated follow-through.
I am building JARVIS to behave less like a dashboard and more like a trusted chief of staff: maintaining context, anticipating what matters, preparing the next move, and acting only within clear human-approved boundaries.
- Claude API
- Node.js
- Plaid
- Gmail
- Google Calendar
- Notion
- MCP
JARVIS / EXECUTIVE AGENTREADY TO ASSIST
OPERATING INTENT
Turn context into clarity, then clarity into action.
01 · UNDERSTANDMaintain contextGoals · people · projects
02 · ANTICIPATESurface what mattersPriorities · risks · next moves
03 · COORDINATEPrepare follow-throughBriefs · drafts · connected tools
04 · LEARNImprove with feedbackPreferences · patterns · outcomes
Human approval gates consequential actions.
Contextmaintains the whole picture
Foresightanticipates needs and risks
Coordinationworks across connected tools
Controlkeeps the human in command
01
Overview
The problem was not missing data. It was fragmented context.
Context
Running a career, two businesses, active product ideas, and a family's financial plan creates information across email, calendars, project systems, and financial accounts.
Goal
Create a reliable briefing layer that surfaces decisions, risks, and commitments without requiring another dashboard to monitor.
My role
Product strategy, system architecture, integration design, prompt and workflow development, testing, and ongoing operation.
Constraint
Personal, communication, and financial data had to remain deliberately scoped, auditable, and separated from the presentation layer.
02
System
A modular pipeline, not a monolithic chatbot.
Each source enters through a bounded connector. The orchestration layer normalizes context, applies schedules and rules, then sends only the relevant information through a model adapter. Outputs are structured for briefings, alerts, and decision support.
Sources
GmailCalendarPlaidNotion
Orchestration
Node.js + MCPScheduling · Context · Rules
Reasoning
Model adapterClaude today · Swappable by design
Outputs
BriefingsAlertsDecision support
Privacy is an architectural requirement.
Connectors are intentionally scoped, sensitive domains remain separated, and the system is designed to minimize the context passed to any model. The public case study shows the pattern—not personal production data.
03
Decisions
Model adapter over lock-in
Reasoning is decoupled from integrations so models can change without rebuilding the data and workflow layers.
Bounded context over unrestricted access
Every integration has a defined purpose and scope. More data is not automatically better data.
Useful automation over novelty
Features earn their place by reducing context switching, improving preparation, or making a decision easier.
04
Outcomes
A working system that changes the shape of the day.
Faster orientation
Daily priorities, preparation needs, and emerging risks arrive in one concise briefing.
Less context switching
Project, calendar, communication, and financial signals are synthesized before action is required.
Better continuity
The persistent knowledge layer keeps decisions and commitments connected across days and projects.
Next
Expand evaluation and observability, strengthen local processing for sensitive workflows, and add more proactive—but still bounded—decision support.
05