Lead Data Scientist (Agentic AI)

Ditto is building the agentic social network — a platform where profiles aren’t static pages, but AI agents that learn from experience, adapt, and help people form meaningful connections.

As an AI-native company, Ditto is designed to operate as a continuously improving intelligence:

• agents learn from real behavior

• systems evolve through feedback

• safety, alignment, and control come first

We believe that systems that learn through interaction will outperform systems trained only on static human data — and unlock a new level of meaningful human connection.

Role Overview

We’re hiring a Lead Data Scientist (Data Engine) to design and own the learning backbone that allows Ditto’s agents to improve safely over time.

You will build systems that:

• capture continuous experience streams, not snapshots

• transform signals into rewards grounded in real outcomes

• feed those rewards back into agents

• prevent drift while still allowing improvement

• help the system reason and plan based on consequences, not guesswork

This role owns the full loop: data → experience → reward → feedback loops → discovering leverage points → adaptation → product outcomes

You are here to build a living system that learns — continuously — and to identify small, high-leverage changes that create outsized impact over time.

What You’ll Build

You will architect the data engine that powers experiential learning:

• experience streams — behavior stitched across long time horizons

• reward streams — derived from actions, outcomes, and environment feedback

• models that capture intent, preference, habit, and change

• evaluation pipelines that measure long-term improvement, not one-off wins

• matchmaking & recommendation signals that uncover hidden compatibility

• systems that let agents plan based on predicted consequences

• experimentation frameworks (A/B tests, bandits, sequential testing)

• drift detection & safety monitors

• guardrails to prevent reward hacking, bias loops, or unintended behaviors

Everything must be auditable, grounded, explainable, repeatable.

Systems Thinking Expectations (Why This Role Is Different)

You will:

• design reinforcing loops that compound value responsibly

• design balancing loops that stabilize trust, fairness, and safety

• identify and avoid system traps (gaming metrics, tragedy-of-the-commons patterns)

• push on leverage points that change behavior — not just parameters

Sometimes the right move is not tuning a metric — it’s redefining the goal.

Must-Have Experience

We want someone who has built systems that learn from experience — not just analyzed history.

• 10+ years in applied ML / data science (production)

• 3+ years building LLM-enabled systems

• built behavioral pipelines that drive real agent / product behavior

• designed feedback & reward loops end-to-end

• hands-on large-scale data engineering

• deep, practical experience with agent frameworks, including:

  • LangGraph (preferred)

  • LangChain

  • or equivalent agent-orchestration frameworks in production

• experience feeding data back into agents to actually change behavior

• strong grounding in:

  • reward shaping

  • value estimation

  • world modeling

  • temporal / TD learning

  • long-horizon feedback loops

If your work stops at insights, this role will feel wrong. If your systems adapt and improve — you’ll thrive here.

Big Pluses

• social graphs, matchmaking, recommendation systems

• trust & safety, anomaly detection, abuse prevention

• causal inference / world-model thinking

• reinforcement learning or TD-style learning

• experience grounding rewards in real outcomes, not proxy metrics

How You’ll Work (AI-Native Collaboration)

You’ll partner closely with:

• AI / NLP — translating signals into agent behavior

• Product — defining success over long time horizons

• Infrastructure — building reliable, observable learning pipelines

• Leadership — aligning learning with business strategy

You won’t just evaluate results. You’ll design how the system learns from them.