Skip to the work

I build AI systems that put your team in the driver’s seat.

I design and build AI systems with your team, from enterprise analytics to full web apps, and teach the people who use them to run them. Looking for my next AI enablement role, full-time or contract.

Jared Schulman

Getting launch teams the right answers from their data.

I consult for the commercial analytics teams at a global pharma company as they prepare to launch a new drug. I help every team, from field staff to leadership, get the answers it needs from launch data, and only the answers it should see.

  1. Governed AI answers for the right people

    I co-wrote the rules the AI assistants follow when someone asks the data a question, so each person gets the answer their role allows.

    Tools. Databricks, Snowflake, Excel, React, Node.js, Microsoft 365 Copilot.

    How I did it

    I worked within the company’s data governance standards and co-wrote three things with the team. First, who sees what: which roles may see which data. Second, a semantic data layer: the approved metric definitions and terms the assistants use, so every answer counts the same way. Third, the test questions and evals that check every answer before it reaches production. The assistants run inside the company’s approved tools.

    Ask as
    Question

    “Who are the top prescribers?”

    BlockedDoctor-level data is not approved for marketing roles. I can show prescribing by specialty instead.

    Same approved definitions for every role. Every answer tested before release.
    Pick a role and a question. Sample data.
  2. Dashboards built around each team’s questions

    I built interactive dashboard prototypes around what each team needed: their key business questions, and the data each role should have to answer them.

    Tools. Claude Code, Snowflake, Excel, React, Node.js.

    How I did it

    I interviewed each team to learn their KPIs, key launch questions and user personas. From that I wrote down what field staff, brand managers and leaders each need to answer, and which data each role should see. Then I designed data visualizations for each kind of stakeholder and built interactive prototypes with Claude Code and React. Every view draws on the same gold-layer data set, so each role gets its own view of one set of numbers.

    Field staff“Who should I see this week?”
    • Dr. A. Rivera14 patients
    • Dr. M. Chen11 patients
    • Dr. P. Okafor9 patients
    Brand manager“Which region is behind?”
    Leader“Are we on plan?”

    Actual Plan

    Sample questions, sample data, made-up names.
  3. Reusable AI skills for analystsIn progress

    I am turning analysts’ manual work into reusable AI skills, so analysts build tools instead of answering the same requests by hand.

    Tools. Databricks, Genie Ontology, Microsoft 365 Copilot, VS Code, SharePoint.

    How I did it

    I studied how analysts handle regular reports and one-off questions, and which requests come back again and again. Then I loaded reference documents and the meaning of the data into Databricks, so Genie agents can find and query the right data for one-off questions. For repeated data work and regular reporting, I am building reusable skills inside the company’s approved Microsoft tools. The design covers the whole organization, so any analyst can pick up a skill instead of writing the same SQL again. The goal is analysts spending their time on insights.

    1. Pull the data
    2. Write the SQL
    3. Check definitions
    4. Build the chart
    5. Write the summary
    Monthly performance summaryOne reusable skill. Same rules every time.
    By handAs a skill
    Sample workflow. Drag the slider.

Before consulting, I spent four and a half years at Capital One, from analyst to business manager, where product and credit analytics meet. See those years

Two AI operating systems

An assistant that knows Brian’s business and works the way he does.

brianlawrence.comBrian Lawrence Web Design and SEO, opens in a new tab

His voice and methods lived in his head, so emails, briefs and reports ate hours every week. I built him a Claude assistant that knows his business, runs client workflows and drafts in his voice.

Tools. Claude Cowork, Gmail, Google Calendar, Google Drive, Trello, Teamwork.

“All I have to do is type in a command in Claude and it’s right there.” Brian Lawrence

Pick a job for Brian’s assistant.Sample

  1. YouBrian asks in plain English.
  2. Business contextPlain files about the business: clients, voice, where things live.
  3. SkillA written procedure for one job. The right one wakes up.
  4. Connected toolsOnly the tools the job needs get touched, each with a limit.
4,989 of his own sent emails taught the assistant how Brian writes.

How it was built

What Brian got

  • Drafts in his voice, for emails, proposals and social posts.
  • His SEO and design methods, used every time, not just when he has the hours.
  • He now builds his own tools, and saves thousands a year on subscriptions he no longer needs.
“All I have to do is type in a command in Claude and it’s right there.”

How I built it

  1. Interviewing Brian

    I asked how he wins clients, reviews a website, talks on calls, and what eats his week.

  2. Collecting his data

    I pulled from the tools he already used: sent email, Trello, Drive and his client list. For his voice, I indexed 4,989 sent emails over ten years and read 443 of them line by line.

  3. Building his files and skills

    Plain files hold his business. Written skills each run one job, with a read or draft-only limit on every connected app. A second AI model reviewed the voice skills before he saw them.

  4. Three days in his office

    I do, we do, you do. We logged 47 issues as he worked. Each fix went in with a test a person runs to confirm it.

How it was built

What KP3 got

  • 48 skills and 21 commands, delivered on KP3’s own laptop.
  • Each client has a written brand file that is read before every job. If it is blank, the job stops and asks.
  • The monthly content job runs in 8 fixed steps and picks up where it left off.
  • Wording gets two separate checks, then a person approves the topics and the words.
  • Sending email and deleting are blocked in the setup itself, not only by instruction. Drafts only.

How I built it

  1. Learning how KP3 works

    A kickoff interview captured their rules and choices before any build.

  2. A brand file per client

    Each client’s voice, themes and channel notes live in a written file. Every job reads it first, and stops to ask if it is blank. The AI never makes up brand facts.

  3. Jobs as written procedures

    The monthly content job runs 8 fixed steps. Two AI reviewers check the words, then the designs, before KP3 approves and keeps editing in Canva.

  4. Test months, then a month with KP3

    I ran full test months on KP3’s own laptop, scored each against fixed criteria, fixed what failed, then ran a month with KP3 at the controls.

Brian, in his own words.

Weekly LinkedIn drafts that start from Colleen’s own answers.

superbloomstrategies.comSuperBloom Strategies, opens in a new tab

Colleen is a fractional CMO, a part-time head of marketing. Client work fills her week, so her best thinking rarely became posts, and AI tools she tried wrote generic posts that didn’t sound like her. Each week this system reads news on her topics and her meeting transcripts, asks her a few pointed questions, and drafts posts from her own answers. Built and tested; Colleen is reviewing it now.

Tools. Claude Cowork, Notion, Tavily, Python checks.

  1. Reads her weekNews on her topics and her recent meeting transcripts.
  2. Asks for her viewQuestions that pull out what she actually thinks.
  3. Saves her wordsHer answers become rows in a Notion idea bank.
  4. Drafts from her ideasHooks and posts built from the idea bank.

Run one week of her content system.Sample

Start demo, answer one question, then save your words. Illustrative content.

weekly-content/SKILL.md, Claude Cowork + Notion

  • NewsTrade piece: mid-size firms are cutting agency retainers
  • Meeting transcriptA prospect asked why their referrals dried up

This week’s question card

  1. Firms are cutting agency retainers. Is that a mistake, or overdue?

    From: news

    Your brain dumpOverdue. Most retainers pay for activity nobody measures.

Nothing yet. Press Start demo to turn news and call notes into a question for you.

Content Ideas a Notion database

Brain dump

Empty. Rows only come from what you said.

Drafting

    You pick what gets drafted.

    Done

    Finished by you.

    Choose your opening line

    1. 8 of 10 checks passed. Missed: takes a clear position, not the obvious take.

    2. 6 of 10 checks passed. Missed: takes a clear position, not the obvious take, hits a real pain, uses a proven hook shape.

    3. 9 of 10 checks passed. Missed: takes a clear position.

    Draft column, LinkedIn

    Firms are cutting agency retainers. Here’s my take.

    Overdue. Most retainers pay for activity nobody measures.

    Illustrative draft. 51 characters in the opening. The empty space is deliberate: your experience belongs here.

    Ideated
    3
    Captured
    1
    Drafted
    1

    Illustrative test log. The real log records the source of each idea when it is connected.

    How it was built

    What Colleen got

    • Her own ideas, easier to use. News and conversation notes become questions she can answer.
    • Drafts inside her Notion. She chooses the idea. Her answer stays with the draft.
    • Room for her voice. Every draft leaves a contribution for her to finish.
    “He has a simple way of explaining complex topics.”

    Colleen Krieger, from her public review of our wider work together.

    How I built it

    1. Learn how Colleen works

      She wanted to speak in her own voice and work in Notion. I split the work in two: finding ideas, and drafting posts.

    2. Turn her material into questions

      The system reads news, her meeting notes and sales calls, and turns them into questions. She answers in her own words, filed in Notion.

    3. Draft only what she picks

      Voice rules come from five of her real posts. Each draft must leave at least one blank for her own story, and an automated check rejects any draft without one.

    4. Test the whole loop

      I ran it in Claude Cowork on test copies of her Notion databases with made-up rows. A second AI model reviewed the drafting rules twice. It is built and tested, waiting on her feedback.

    It reads a business’s website, then asks only for what is missing.

    instantAIpage. An AI web app built for an SEO agency to use with its clients.

    instantAIpage gives a business an official page that AI search tools like ChatGPT can find and quote accurately. It reads the owner’s website, spots what’s missing, and asks tailored questions that draw out what sets them apart from competitors. Then it writes the page and its structured data.

    Tools. Firecrawl, OpenRouter, Claude Haiku 4.5, Claude Sonnet 4.5, OpenAI embeddings, Supabase pgvector, Next.js, Vercel.

    Run it on a sample venue’s website.Sample

    1. Start
    2. Read the site
    3. Check the facts
    4. Fill the gaps
    5. Your page

    Press Start demo, check the facts, then answer what is missing. Illustrative business.

    • Home
    • About
    • Weddings
    • Gallery
    • Kind words
    • Contact

    HomeFernhollow Barn. A restored dairy barn for weddings in Alder Creek.

    WeddingsCeremonies in the orchard, dinner for up to 160 in the barn, and a bridal loft upstairs.

    AboutRun by the Okafor family since 2014.

    Kind wordsReview text is omitted in this illustrative demo.

    ContactCall 555 0142 or write hello@fernhollowbarn.example

    JSON-LD schema, built by code

    This is the file a search engine or AI assistant reads. You don’t need to read it. Notice that it only holds facts from the site and your answers.

    1. {
    2. "@context": "https://schema.org",
    3. "@type": ["EventVenue", "LocalBusiness"],
    4. "name": "Fernhollow Barn",
    5. "url": "https://fernhollowbarn.example",
    6. "telephone": "555 0142",
    7. "foundingDate": "2014",
    8. "address": { "@type": "PostalAddress",
    9. "addressLocality": "Alder Creek" },
    10. "areaServed": "Alder Creek and the Pell Valley"
    11. }

    Later, someone asks an AI assistant

    Who does barn weddings near Alder Creek for about 150 people?

    An illustration. No tool can promise that an assistant will cite a page. A clear, factual page, organized so software can read it too, gives it something accurate to cite. Touch a phrase to see where it came from.

    What we found

    Your website already knows most of the answers. Watch them land here.

    Business name
    Type
    Wedding venue
    Location
    Alder Creek
    Services
    Orchard ceremonies, barn dinners for up to 160, bridal loft
    Founded
    2014, family run
    Testimonials
    Not reproduced in this demo
    Contact
    555 0142, hello@fernhollowbarn.example
    Service areas
    Not on the site
    What sets you apart
    Not on the site
    Ideal client
    Not on the site
    AI reference pageShortened sample

    Official information about Fernhollow Barn

    Basic information

    Fernhollow Barn is a wedding venue in Alder Creek. It has been family run since 2014.

    Services

    The venue hosts ceremonies in its orchard and dinners for up to 160 guests in the barn. A bridal loft is upstairs.

    Service areas

    Fernhollow Barn serves Alder Creek and the Pell Valley.

    What sets it apart

    In the owner’s words: “One wedding per weekend, so nobody is rushed out.”

    Ideal fit

    Who it suits, in the owner’s words: “Couples planning 80 to 160 guests who want it outdoors.”

    Connect

    Phone 555 0142. Email hello@fernhollowbarn.example.

    About this page: facts come from the business’s website and its owner. Sample content.

    How it was built

    What the owners got

    • Start with an existing site. Owners check what the tool found instead of repeating it.
    • A page and its supporting files. HTML, plain text and structured facts are ready to download.
    • A review step for the agency. The owner can send the finished package for review.

    How I built it

    1. Start with the friction

      The agency’s wedding-industry clients faced a long questionnaire. The tool starts with the website they already have.

    2. Read the useful pages

      Firecrawl reads the site. Claude Haiku 4.5, called through OpenRouter, pulls out the business facts and likely answers.

    3. Let the owner check

      The owner reviews the facts and corrects them. The system asks only for information it could not find. Its question bank is stored as OpenAI embeddings in Supabase (pgvector), and embedding search picks the right question for each gap while skipping ones the chat already asked.

    4. Write the page, build the facts

      Claude Sonnet 4.5 writes the reference page. Plain code builds the structured file from the reviewed facts. The owner downloads it all, or sends it to the agency for review.

    5. Test it and take feedback

      41 automated test files check the code. The agency owner reviewed the mockups for the latest phase, and his feedback was logged before I built.

    One sample venue goes through the tool, then a look under the hood. The business names are made up.

    A weekly board of companies ready to buy, built by AI agents and approved by a person.

    clientmine.aiClientMine, opens in a new tab

    ClientMine is a web app that tells a firm, every week, which companies are ready to buy. It checks two things together: signals of buying readiness, like hiring abroad, new funding or SEC filings, and fit, meaning the right type of company with a clear decision maker. Each company comes with a tailored approach, and a person approves the board before the firm sees it. Its first paying client is an immigration law firm.

    Tools. Claude Code agents, Python scraping, Firecrawl, SEC EDGAR API, PredictLeads API, Supabase, Next.js, Vercel.

    The technical write-up

    Run a weekly scan. Watch which steps are AI agents and which are plain code.Sample

    1. Find companies with a buying triggerAIScout agent
    2. Read each company’s website onceCodeCode
    3. Check the gates, score the fitAIScorer agent
    4. Research the best, up to 6 a weekAIEnricher agent
    5. Write an approach for eachAIEnricher agent
    6. Final score and tierCodeCode, no AI
    7. You approve the boardYouA person
    8. Publish to the firm’s boardCodeCode
    9. Recheck last week’s companiesAIMonitor agent, alongside

    This week’s scan hasn’t run yet.

    Last week’s board is still live for the firm. Press Run it to watch the agents build the next one.

    1. Larkspur Biotech

      Raised a Series B in Germany

      Site readTier A

      Head of People. Email found.

      First moveCongratulate her on the raise. Ask how the German team will move over.

    2. Tidewell Analytics

      Opening a U.S. office

      Site readTier A

      COO. Email found.

      First moveAsk who handles visas for the U.S. launch team.

    3. Oakmere Foods

      Local bakery, no hiring

      Site read

      Too small
    4. Brightline Freight

      Hiring 12 engineers abroad

      Site readTier B

      VP Engineering. On LinkedIn.

      First moveAsk about timing for the 12 hires abroad.

    5. Quillon Software

      New VP of People

      Site readTier B

    6. Marrow & Pine

      A law firm itself

      Site read

      Wrong industry
    7. Halcyon Labs

      Expanding to Texas

      Site readTier B

    8. Vessel Robotics

      No named leader to reach

      Site read

      No way in

    Hot ListNew this week

    Acting now

      Worth building

        How it was built

        What the firm got

        • A board of companies to call, each with why now, a fit score, who to contact and a first move.
        • Research spent where it counts. Cheap checks drop poor fits first, and deep research goes to at most 6 companies a week.
        • Nothing goes live without a person. The new board waits for approval, and a failed scan never replaces last week’s board.

        How I built it

        1. Split judgment from mechanics

          AI agents do the reading and judging: finding companies, scoring fit, researching contacts and watching for changes. Plain Python runs the order, the rules, the final score and the saving.

        2. Hand work over in files

          Each agent writes its results to a file in that week’s folder. The next step reads it, and a checker rejects any file that breaks the agreed shape.

        3. Put a person at the gate

          The finished board waits for approval, tied to that exact board. The database write refuses to run without it.

        4. Ship it as a product

          A Next.js app on Vercel, Supabase for each firm’s data and board, and PredictLeads for hiring and funding signals. The first paying firm joined in July 2026.

        5. Test every rule

          About 170 automated test files cover the plain-code steps, the file checks, the gates and the database write. A failed scan never replaces last week’s board.

        The agents

        • Scout finds companies showing a buying trigger.
        • Scorer checks the gates and scores fit.
        • Enricher researches the best companies, finds who to contact and writes an approach for each: a first move, a draft first message and questions for the call.
        • Monitor rechecks earlier companies for changes, alongside the rest.

        Each agent is a Claude Code agent following a written procedure. Everything between them is plain Python.

        From analytics to operations to building AI with teams.

        Understand the data. Learn how operations run. Then build with the people who use it.

        Education
        University of Virginia, 2017BA Cognitive Science and BA Psychology
        Languages
        SpanishProfessional working proficiency
        PortugueseBasic conversational
        1. Aug 2017 to Feb 2022. McLean, Virginia.

          Capital One

          Analytics, product and risk.

          What I did at this stage
          • Owned yield management analytics for the largest card segment, building fee and interest resiliency models that supported a $40 million credit decision.
          • Automated leadership dashboards on fees, interest yield and waiver rates in Databricks and Tableau with SQL, Python and Spark.
          • Launched a financial-hardship relief partnership during COVID, from first conversation through legal, risk, compliance, cybersecurity and finance reviews, and built the business case that won long-term investment from senior leaders.
          • Led product strategy for a spend-tracking feature in the mobile app, through beta testing and user research, and ran A/B tests for new card accounts.
          More detail on my resume (PDF)
        2. Mar 2022 to May 2025. Washington, D.C. and South America.

          Hospitality Startup Venture Research

          Hostel venture research. The first AI builds.

          What I did at this stage

          I researched the hostel business from inside the industry to inform a go or no-go on a startup investment, working hands-on up to general manager of a 400+ capacity, 40+ staff hospitality operation. I built a web scraper with early ChatGPT, Playwright and Selenium to track hostel occupancy by city, which informed a no-go decision. Before that I taught English and volunteered in hostels across Colombia, Peru, Bolivia, Argentina and Brazil (Mar 2022 to Feb 2023), reaching professional working proficiency in Spanish.

        3. May 2025 to present. Remote and Arlington, Virginia.

          Now

          Two consulting roles, side by side.

          Global pharma company

          AI Enablement Consultant (Contractor)

          Jun 2026 to present

          Dashboards for each role, governed AI answers and reusable AI skills for the commercial analytics teams, ahead of a drug launch.

          OtterMatic

          Founder & AI Systems Builder

          May 2025 to present

          Custom AI workspaces for professional-services firms, training the people who run them, and products like ClientMine and instantAIpage.

        Let’s build your team’s next chapter.

        Hiring for AI enablement? I’m looking for a full-time role or a contract where I build with your team and help them take over.

        Three ways to work with me: an AI enablement role on your team, an embedded consulting contract, or a defined build with training for your staff.

        jaredlschulman@gmail.com
        Jared Schulman
        How this portfolio was built
        How it works
        The demos on these pages run in your browser, written in TypeScript with SVG and Canvas. One small helper handles the spring motion and the puck. Approval pucks work with a click, Enter or Space. Sliders also work with the arrow keys. Reduced-motion settings are respected, and the pages still read with JavaScript off.
        What the demos are
        Simplified versions of the real systems. They send nothing and save nothing.

        Jared Schulman, Arlington, Virginia. No tracking, no cookies.