Back

Case Study

Jarvis

My personal AI operating system. Built to close the gap between knowing what to do and actually doing it.

Next.jsClaude · OpenAI · PerplexityMem036+ Subagents75+ Skills50+ Integrations

The Problem

I was running everything in my head.

I'd finish a sales call with real momentum. Good discovery, clear pain expressed by the prospect, next steps outlined. Two weeks later, I'm digging through a Fireflies transcript trying to remember what they said about their procurement process. The context was always somewhere... just never right in front of me.

Email was a morning killer. I'd open my inbox to check one thing and 40 minutes later I'd still be in there. Not because the emails were hard. Because triage is mental work. What needs a reply? What's junk? What's a deal I should actually move on? Doing that manually every morning is a tax on the day before it even starts.

Content was the worst. I had ideas constantly. I'd write something, save it somewhere, and it would just... die there. The newsletter, the LinkedIn posts, the brand... all inconsistent because production and distribution were both manual and both competing with a full-time quota.

I was running a W2 job, three side brands, a stock portfolio, and a job search at the same time. Each one in a different tool. Each tool with a different login, a different mental model, a different context to rebuild every time you open it.

The problem was never effort. The problem was friction. And friction compounds.

What It Does

A morning with Jarvis.

Jarvis lives in two places. There's the dashboard: a custom-built interface I designed for myself, connected to every part of my digital life. And there's the chat window, where I give commands and ask questions. Both are always running.

What the dashboard surfaces

  • Email triage: inbox already sorted into junk, FYI, and action. I see what matters before I've had coffee.
  • Calendar, todo list, and revenue goals: where I am today, what's blocking me, and how far I am from my number.
  • Full pipeline view: every open deal, stage, score, and last touch. The ones that need my attention surface first.
  • Prospect intelligence: a live signal feed watches for news, job changes, funding announcements, and buying signals tied to my ICP. No manual research.
  • Reddit scraper: surfaces threads in my domain where I should be adding to the conversation, not just broadcasting.
  • Content feed: a curated news stream tied to my domain, with a content idea generator that turns what I'm reading into posts worth writing.
  • Full CRM: every deal, every note, every call transcript from Fireflies, every email, every text message — all in one place. The context is always there.
  • Portfolio tracker: live account balances and positions, including a separate trading account Jarvis manages autonomously using a swing trader skill I built for him.
  • Demo assistant: listens to live sales calls and checks off what I've covered in real time. MEDDPICC qualification, feature coverage, objections. If I miss something, it surfaces it.
  • And yes... a recipe section. I love to cook.

What the chat executes

  • “Draft a reply to that prospect email” ... done in 30 seconds, in your voice.
  • “Score this deal with MEDDPICC” ... pulls deal context, scores every dimension, flags the gaps.
  • “Build me a lead list of staffing firms in the Southeast” ... scrapes Google Maps, Yelp, Yellow Pages, drops it into a Sheet.
  • “Place a trade on that ticker” ... checks the balance, executes the order, confirms the fill.
  • “Text my contact at that account” ... drafts the message, shows me first, sends on approval.

What Jarvis runs automatically

  • Morning briefing: a cron job that fires every morning and prepares me for the day. What's on the calendar, what deals need movement, what emails need a reply. I don't have to go looking — it's already there.
  • Afternoon briefing: a second cron that resets the lens for the back half of the day. What changed, what's due before EOD, what do I still need to close out.
  • Pre-call briefing: 30 minutes before go time, Jarvis delivers everything I need on the person I'm about to talk to. Who they are, where they live, where they went to school, personal details surfaced in prior calls to help build rapport, where we are in the deal, and exactly which MEDDPICC gaps I still need to uncover. I show up prepared every time.
  • Jarvis brain file: a running system of what Jarvis thinks I want. He's been given the ability to make certain decisions on his own, without my input — based on patterns, preferences, and prior context he's accumulated over time.

You show up, ask one question or give one command, and Jarvis either answers it or executes it. No context-switching. No five tabs.

The Build

What was actually hard.

The core system took a few weeks. It's still being built. Jarvis is a living operating system, not a finished product.

Voice fidelity.

Getting Jarvis to write as me, not as a generic AI assistant, meant building a full knowledge base from scratch: voice guides, example posts, anti-patterns, things I'd never say. Generic LLM output is obvious. Getting it to sound like you, at your best... takes real work.

Cross-system context.

Jarvis pulls live data from Gmail, Calendar, the CRM, Schwab, Twilio, Fireflies, Apollo, Tasks, and Drive all at once. Making all of that coherent in a single prompt, without hallucinating or mixing contexts, was the core engineering problem.

The CRM layer.

MEDDPICC scoring, deal notes, transcript sync from Fireflies. Building a lightweight CRM that reflects how I actually sell, not how Salesforce thinks I should, required real customization.

Trust.

The system only works if the data in it is accurate and current. Getting the inputs right, deal stages, scoring, notes, is the ongoing discipline. You can't automate your way out of bad inputs.

The Stack

Everything talks to one interface.

Frontend / runtimeNext.js
AI providersClaude, OpenAI, and Perplexity — all three depending on the task
MemoryMem0, persistent memory across conversations
CRMCustom-built, file-based (markdown + JSON), Fireflies for transcript sync
EmailGmail API, triage, drafting, sending
CalendarGoogle Calendar API
BrokerageCharles Schwab Developer API, live portfolio and orders
SMSTwilio
ProspectingApollo.io, Hunter.io
Web intelligenceFirecrawl, Jina Reader
Lead listsOutscraper, Google Maps, Yelp, Yellow Pages
Content pipelineCustom markdown-based queue, Google Drive
Job searchJSearch API, LinkedIn, Indeed, Glassdoor, ZipRecruiter
Stock dataPolygon.io
Task managementGoogle Tasks API
Knowledge baseMarkdown files, voice guides, deal playbooks, frameworks

The Impact

What actually changed.

Speed.

Tasks that used to take 20 to 30 minutes, drafting a follow-up, scoring a deal, building a prospect list, now take under 2 minutes.

Coverage.

W2 job, 3 side brands, a content engine, a stock portfolio, and a job search running at the same time. Without a system, things fall. With Jarvis, the pipeline gets checked, emails get answered, content gets drafted... even when your focus is somewhere else entirely.

Deal quality.

MEDDPICC scoring surfaces the qualification gaps before they cost you. Jarvis flags what's missing while there's still time to close it — not after the deal goes quiet.

No rebuilding context.

You don't have to remember what's in each deal, what you said last week, or what's due. You ask. You get the answer. You move.

The honest version: the system is only as good as the data you feed it. Deals with no notes get no coaching. Jarvis amplifies the discipline you bring... it doesn't replace it.