hussh 🤫

Founding Public Policy Data Scientist

New York, NY

Full-time

Published Expires (UTC)

Your next opportunity

Create a job alert

Receive new jobs matching these criteria.

Data ScientistsNew York, NY · Within 15.5 mi · Data Science & Analytics, Full-Time Jobs
Free for job seekers. Always.
You’re in control

Manage your alerts or unsubscribe at any time.

Manage job alerts

About this opportunity

The mission

The most valuable asset in the world is what you search, buy, earn, owe, and are. It is collected, traded, and used by companies you have never met, under terms you never negotiated, for purposes you never approved. Hussh exists to end that.

Every generation writes the rules for a new kind of property. Ours is information — and this time, the owner should be the person. We are campaigning to establish information rights as a new bill of rights: every American's right to their own information — to know what is held about them, to control it completely, and to permit or refuse its commercial use, which may happen only with their explicit consent, and in which they share directly.

It is the foundation under everything we build: Private Agent One; the 🤫 money app; a permanent-capital asset manager. Our products are where the right becomes real: personal models of a person — their world, their knowledge, their judgment — built from their information, under their control, belonging to the person they describe.

Everything we share with lawmakers is public, in one folder: the American Information Rights Act, the constitutional path, the case, the two-page summary and the full brief, our one-pager, and our press release.

https://drive.google.com/drive/folders/1FHTQIyFsR5ealf2QCRTZ_zvRDNSzBtFo

The role

Founding Public Policy Data Scientist — Hussh's first dedicated data hire.

Be precise about what this is: a data science and data engineering role. Policy is the domain; data is the craft. You will build the data foundation for a constitutional campaign and a company: pipelines that turn the world's legislation into structured knowledge, the models of people our products run on. You will model the public record — lawmakers acting in public — and, only with consent, private people; design to the stricter regime.

What you'll do

- Build production data pipelines end to end over the primary sources of law: Congress, state legislatures, the Federal Register, agency rulemaking, and their equivalents in India, the UAE, and Europe — systems that run daily without you watching them.

- The difficulty is breadth, not volume: fifty state legislatures with fifty formats, bulk files beside scanned PDFs that need OCR, bills in five versions whose differences are the story, sources with no stable identifiers and no changelog. A source's terms of use are part of its provenance: a company built on consent does not scrape carelessly.

- Build the pipelines behind our models of people — world models, knowledge models, and supermodels: per-person models built from information the person has consented to share, kept current, and put to work for better communication, better connections, and richer commerce — with the person, never a platform, in control. Own the refresh and evaluation pipelines that keep those models true: incremental updates, staleness you can measure, contradiction detection, sampled human audits of factual accuracy.

- Architect consent so it is a property of the data, not a promise about it: every fact carries its source, timestamp, confidence, and the consent it rests on; consent and purpose are enforced as queryable constraints at read time; deletion propagates to derived data — features, embeddings, summaries; access is scoped and logged. When a person withdraws consent, their data and everything derived from it is invalidated within a stated time.

- Turn unstructured legal text into structured, queryable data products: who is proposing what, where it stands, what it changes, whom it affects. LLM-assisted extraction where it earns its place — with versioned gold sets, per-field precision and recall you can state, human review loops with measured agreement, pinned model and prompt versions, and a cost per document you can defend. Not vibes; numbers.

- Build the knowledge layer on the people and institutions who decide — lawmakers, committees, agencies, universities — maintained as data with sources — and own the campaign's measurement.

- Own data quality as a discipline: schemas, tests, lineage, and data contracts with change detection on every external source, because a legislature's website will change its format the week you depend on it. Any number this company cites traces to its source.

- Build the internal tools that let a founding team produce evidence at the standard of an institution, automate your own repetitive work out of existence, and work directly with the founder, whose priorities your analysis will set.

What success looks like

In 90 days: when anyone at Hussh asks what is happening in policy, the answer comes from your data, not from memory: a live legislative corpus, campaign measurement on your pipelines, every headline number reproducible from source.

In your first year: the information-rights effort runs on infrastructure you built — the campaign's reference dataset, the evidence behind every lawmaker briefing, and consent enforced as schema, not policy. The personal models our products run on are fed by your pipelines, each traceable to consent.

What you bring

Technical depth is the price of entry; we test for it seriously, and show you the same rigor we ask for.

- Deep data science: statistics you can defend — experimental design, causal inference, multiple-comparison discipline, base rates before anecdotes, and uncertainty stated honestly. Most campaign measurement is small-n, clustered, and confounded; you have told a stakeholder "this cannot be measured yet" and been right.

- Production-grade engineering: typed, tested Python and SQL deep enough to be boring — query plans you have read and fixed. pandas and scikit-learn are table stakes; the headline is software other people can run, and pipelines that survived the real failure modes: backfills, late-arriving data, silent schema changes, duplicates, reruns that must be idempotent.

- Entity resolution is core craft here, not a nice-to-have: you have matched people and documents across sources that share no identifier; you know blocking and pairwise matching, and can state the precision/recall tradeoff of a matcher you shipped — and why, for a knowledge base cited to lawmakers, precision wins.

- You have built per-entity models from messy, disagreeing sources before — identity graphs, knowledge bases, research profiles — and you think in valid time as well as transaction time: what was true, when it was true, and when we learned it are three different facts in your schemas. Permission-filtered retrieval (search and RAG scoped per record) is a strong plus.

- Serious AI practice: you use AI tools daily — coding agents, LLM pipelines, automated extraction — and you know where AI ends and verification begins. Bring examples.

- Multilingual text as data: Arabic, Hindi, and European languages are in the corpus, cross-lingual matching included.

- Writing and domain: brief, exact, honest analysis, and fluency in technology policy (privacy, data rights, AI governance) or the demonstrated ability to get fluent fast. The domain can be learned. The data craft cannot be faked.

- Conviction: the idea at the center of this company — that a person should hold complete, consent-based control of their information and participate directly in the information economy it creates — is one you have already arrived at on your own. If the mission in the folder above does not move you, this is not your seat.

Both steps matter: the email is where your answers to the three questions below are read.

Answer three questions. They are the application.

1. Show us the hardest data system you have built and run in production: how many sources and how much data it moved, the worst incident it survived, and how you proved it correct — reconciliation, checksums, gold sets, backtests. What would you rebuild differently? Links welcome.

2. How do you use AI in your daily data work? Give one concrete example where AI changed what you shipped — and one where you overruled it: what the AI produced, how you caught the error, and what check now exists because of it.

3. In a few sentences: why should a person — not a platform — control their information and its commercial use? Then the real question. Pick one public figure. Sketch the model you would build in 90 days: your sources; your schema for a fact (its provenance, confidence, and consent or public-record basis); how the model stays current; how you would measure its accuracy; and, for a private person who withdraws consent, what happens to their data and everything derived from it. Two pages maximum.

We are remote friendly, and we welcome candidates from all over the world to consider coming and working for us in this role.

Terms: full-time. Posted in New York — the largest data-scientist pool in the country, where technology, finance, and policy meet — with our home base at the Hussh Garage in Kirkland, Washington. Work from New York, Washington, DC, Chicago, or Kirkland; if you relocate to join us, we will help. Compensation is competitive for the experience, and includes equity under the company's equity plan, subject to board approval.

Benefits — we take care of our people the way the best employers in America do:

- Health: comprehensive medical, dental, and vision for you and your family — premiums paid by the company.

- Retirement: 401(k) with a company match; and, as an owner, the chance to invest alongside the house.

- Time: generous paid time off, paid holidays, and sick leave.

- Family: paid parental leave for birth, adoption, and bonding; fertility and adoption support.

- Security: life and disability insurance, and a survivor benefit for your family.

- Growth: a yearly learning budget, and the best AI tools in the world, on us.

- Relocation: we carry the cost of your move.

Benefits

  • Disability insurance

- Security: life and disability insurance, and a survivor benefit for your family.

  • Paid holidays
  • Paid time off

- Time: generous paid time off, paid holidays, and sick leave.

  • Paid parental leave

- Family: paid parental leave for birth, adoption, and bonding; fertility and adoption support.

  • Employer retirement match

- Retirement: 401(k) with a company match; and, as an owner, the chance to invest alongside the house.

Your next step

Interested in this role?

Review the details, then choose the resume you want to share.

Apply for this job Free for job seekers. Always.

Share this job

Keep your search safe

Employers make hiring decisions. Never pay someone to secure a job.

Job search safety
About the employer

hussh 🤫

Represent this company?Claim this profile
Your preferred location

Where would you like to work?

Choose a city, a state, or United States to search across the country.

Location is requested only when you choose this option. Precise coordinates are not stored.
JobbolJobbol

Your privacy. Your choice.

Choose optional cookies and data uses in this browser. Closing this window does not give consent. Limited Google measurement without optional cookies may run before your choice; recorded refusals and browser privacy signals are respected.

Essential services

Sign-in, secure forms, abuse prevention, language and the city or theme you explicitly select. These functions are needed to provide the service you request.

Always active

Do Not Sell or Share My Personal Information

Jobbol does not sell resumes or use your personal information for cross-context behavioral advertising. This protection is active for everyone. You can also turn off all optional processing below without signing in.

Protection active

You can change your choice at any time from the footer. Choices last up to 180 days in this browser. Account restrictions also apply when we can identify your signed-in account.

Notice version: 2026-11-12.ads1

Quick view