Product manager and founder
I build AI products for people the market skips. I run TheThroughLine.AI and I am building TheDirectLine.AI at the Harvard Innovation Labs.
Recruiting for Summer 2027 product roles
Two products, one system
TheThroughLine.AI is the free media layer. TheDirectLine.AI is the practice layer. Each one feeds the other.
Live todayUnderstand today. Shape tomorrow.
Free articles, research, and short-form video with insight from leaders in business, technology, and policy. Every piece includes a hands-on AI exercise.
In development at Harvard Innovation LabsThink first. Then multiply.
Teaches students to use AI as a multiplier on their own thinking, not a replacement for it. Teachers set drills, see every answer, and approve what students see.
Platform counts load live. Some are manually verified when a platform blocks requests.
How I use AI as a PM
I run the same six questions on every product. AI does the research, clustering, and scoring. I make the calls. Each card shows the decision and where AI helped.
Users are learners. First buyers are program directors at college-access nonprofits and community colleges. School districts are the long-term buyer. Beachhead: Texas.
Where AI helped. Claude clustered TikTok Studio audience data and Texas field notes into two user segments and one buyer segment.
60% of Black teens and 62% of Hispanic teens use chatbots for schoolwork. Only 15% of each group rank as highly AI-ready, versus 32% of White teens.
Where AI helped. I used AI to pull 27 sources, then logged and verified every figure. v0.2 cut four numbers I could not trace to a primary source.
67% of low-poverty districts trained teachers on AI in fall 2024, versus 39% of high-poverty districts. Students stop at asking for answers. Each pain point stays a hypothesis until interviews confirm it.
Where AI helped. Claude scores each assumption after every interview batch so I see which bets are getting stronger or weaker.
Students answer on their own while AI stays locked. Then they write up to 3 prompts, a coach scores clarity, context, and constraints, and they write the final answer and a reflection.
Where AI helped. Runs on local open-source models, not a wrapper on a hosted API. Token efficient today. We scale compute as pilots prove the use cases and features school districts will pay for.
Pilot target: 60% of enrolled learners. Supporting metrics: +25% blind-scored skill gain, 3+ revisions per artifact, 70% active in week 6 of 8.
Where AI helped. The north star counts finished work, not logins. It rewards depth, which is the problem I am solving.
AI never writes a student's final answer. Feedback must match educator grading on 85% of a 100-attempt eval set with zero fabricated facts. Student data stays at a display name and school email.
Where AI helped. No challenge ships until it passes the eval set. Zero-shot AI grading agrees poorly with human graders, so I test it first.
Measuring success
The journey runs from a 60-second video to a paid seat. Today the weak link is the return. About 3% of viewers come back.
| Step | Metric | Today | Week 8 goal |
|---|---|---|---|
| Discover | Weekly TikTok views | 7,400 | 20,000 |
| Return | Returning viewers | About 3% | 15% |
| Engage | Comments per week | 17 | 150 |
| Read | New Substack subscribers per week | Tracking | 50 |
| Ask | Correct citations on a 75-question test | Not built | 90% |
| Pay | Students and schools on a priced waitlist | 0 | 20 |
Baselines: TikTok Studio, Week 4 pull, Oct. 4, 2026. Goals: ThroughLine Week 5 to 8 growth plan.
Bet
Named series and a comment funnel.
Goal: 3% to 15% returning
Bet
Ask Us Anything with 25 to 50 Texas students on their own wifi.
Goal: Preference on real connections
Bet
15 to 20 student and 5 school interviews, plus a priced waitlist.
Goal: 20 on the waitlist
Track record
Here is what I did, and the rule I took from it into TheThroughLine.AI and TheDirectLine.AI.
Summer 2026
MBA Product Marketing Manager Intern, AI Agents
What I did
Built go-to-market positioning for AI agents across Accounts Payable, Accounts Receivable, and AI Finance HelpDesk. Led alignment across Product, Sales, and Customer Success to launch an economic toolkit that quantifies cost savings.
Product lesson
A model is not a product until a buyer can see the outcome. Agents also fail without company context. That is the same gap TheDirectLine.AI closes for students.
2023 to 2025
Associate, Global Credit & Markets Operations
What I did
Led product strategy and launch for CheckMate, an AI-assisted deal pipeline platform. Ran research with 50+ portfolio managers and analysts and drove adoption across a $70B+ credit portfolio. Automated fund onboarding and KYC checks, lifting reporting accuracy 25%.
Product lesson
Start with the user's workflow, not the feature list. Accuracy is a guardrail. In regulated work, one wrong number costs more than a slow one.
2021 to 2023
Senior Analyst, Global Banking & Markets
What I did
Built the business case for a firmwide internal platform and partnered with Product and Engineering on rollout. Drove a 30% engagement lift across 40,000+ employees through enablement and storytelling.
Product lesson
Shipping is half the job. Adoption is the other half, and it needs its own plan, owners, and metric.
MBA, Tech Strategy, Expected 2028
“Wise, decisive leaders who better the world through business.”
Product Management in Tech and AI and Consultative Decision-Making. I make a call with incomplete data, name the risk, and set the metric that tells me if I was wrong.
MPA, AI Policy and Governance, Expected 2028
“Ask what you can do.”
Policy shapes what AI products can ship. FERPA, data governance, and trust become product requirements on day one, not a review at the end.
The operating system
TheDirectLine.AI runs and develops on local open-source models to stay token efficient. It is not a wrapper on a hosted API. Compute scales as we prove the use cases and features that fulfill willingness to pay for school districts. I build the tools, plans, and playbooks myself. Read the originals below.
Classroom AI practice tool running on local open-source models. Token efficient by design, with compute scaling as pilots prove what school districts will pay for. Teacher dashboard, prompt scoring, and reflection.
A 10-step operating loop: monitor, diagnose, pace, allocate, optimize, plan, measure, act, communicate. Claude runs the daily and weekly tasks.
Every test names a primary metric and a guardrail. Success means the primary metric rises 20% or more and the guardrail holds.
Problem, segments, assumptions, three approaches, and a 90-day roadmap for Harvard Innovation Labs and MIT Media Lab mentors.
Loading document…
Proof of work
Followers by channel and the latest writing from TheThroughLine.AI.
Channel audience
Live or manually verified
TikTok
163
Followers
2.3K likes
YouTube
4
Followers
Substack
11
Followers, manually verified
34
Followers, manually verified
These are early numbers. Counts labeled manually verified may be older than the latest platform data. Other counts are requested when the page loads.
Live evidence, Substack
Alphabet had its worst day ever, then rallied with Big Tech after the midterms. Both moves come down to one question: who pays for an AI answer?
Voters rejected data centers town by town. Now a new Congress inherits the hardest question in tech: who pays for the AI buildout?
For years, EVs cost more up front and saved money later. That rule just broke on BMW’s 3 Series. How to compare the full cost of your first car.
Founder Building, TheThroughLine.AI
A 15-page paper on transformer architecture, its effects on business and policy, and why you still need to check the output of AI tools.
Loading paper…
By Everrett V. Harris, TheThroughLine.AI. Read and navigate all 15 pages here, or open the full paper in a new tab.
Summer 2027
