Full Stack AI Engineer, Creative Intelligence

We are a legal-intake advertising operation with a unique internal platform that monitors every ad campaign by measuring cost per signed case. Our system uses a multimodal AI model to analyze videos, generating timestamped transcripts and scene breakdowns. We map viewer retention to these scenes, tag each creative with detailed attributes such as hook pattern, spokesperson, claim type, emotional arc, and offer framing, and link all data to signed cases from our intake CRM. At the core is the Brain, an AI that diagnoses why a creative may be underperforming, identifies where viewers drop off, suggests script improvements, and tracks predictions against actual outcomes. This platform is live and evolving, and your role will be to enhance its accuracy and reliability while managing the entire system.

Key Responsibilities
- Own the platform end to end: Build and maintain the application used daily by strategists, media buyers, and editors. Manage integrations with Meta, Google, TikTok, and CRM systems, as well as nightly batch jobs.
- Validate the Brain’s performance: Develop an evaluation layer to measure diagnosis quality against expert judgment and prediction accuracy against actual signed cases. Improve confidence calibration and create clear metrics for leadership.
- Improve attribute extraction: Increase consistency in tagging subjective creative elements to meet an 85% agreement threshold between AI and human tags, enabling these attributes to be fully integrated into performance analytics.
- Apply robust statistics: Address data sparsity by implementing partial pooling techniques so the Brain can make reliable probabilistic statements even with limited data.
- Refine diagnosis methodology: Transition from a single large AI prompt to a modular system with retrieval of past briefs, claim verification, and appropriate handling of limited evidence.
- Implement causal testing: Design and run creative holdout experiments to distinguish correlation from causation in campaign performance improvements.

Required Qualifications
- Full-stack TypeScript expertise: Proficient in Next.js App Router and React (Next 16, React 19), including server components, server actions, and streaming responses. Able to deliver end-to-end features from database schema to user interface with a focus on user experience.
- Strong backend and database skills: Experience with PostgreSQL, analytical SQL (window functions, CTEs, complex aggregations), schema design, background jobs, pipelines, and containerized production environments.
- Production AI experience: Proven track record shipping LLM-powered features with structured JSON output, schema validation (e.g., Zod), tool calling, retries, cost management, and retrieval-based grounding. Able to measure and report AI accuracy using golden sets, LLM-as-judge, or regression tests.
- Statistical understanding: Basic grasp of why small sample sizes yield unreliable conversion rates; deeper statistical knowledge is a strong advantage.

Preferred Qualifications and Benefits
- Advanced statistics for sparse data and experimentation: Familiarity with hierarchical or empirical Bayes methods, shrinkage, partial pooling, and experimentation design. Experience with beta-binomial or gamma-Poisson models is a plus.
- Vector search technologies: Experience with pgvector, HNSW indexes, and cosine similarity for embedding-based retrieval within PostgreSQL.
- Model-agnostic AI workflows: Comfortable routing requests across multiple LLM providers such as Gemini, GPT, Claude, and Grok to optimize cost and performance.
- Ad platform API integration: Knowledge of Meta Marketing API, Google Ads API, TikTok Marketing API, and conversion tracking methods.
- Creative team collaboration: Understanding of creative workflows and the reasons behind their skepticism of dashboards and analytics tools.

Our technology stack is fully TypeScript-based, running on Bun with Next.js 16, React 19, Tailwind, shadcn, TanStack Query and Table. We use PostgreSQL with pgvector, Drizzle ORM, Supabase authentication, OpenRouter for model access, and deploy via Docker on Railway with scheduled nightly jobs. Familiarity with any part of this stack is beneficial but not mandatory.

We do not require a marketing background or advanced degrees; what matters most is your ability to ship measurable, reliable AI-driven features. Our team is small and agile, with direct access to decision-makers. You will work across the entire stack, shipping changes that include schema, queries, server actions, access control, and UI, all validated against real data. The Brain proposes insights, but humans make the final decisions, and every AI claim must be traceable and verifiable.

If you are excited to build a cutting-edge AI platform that empowers creative strategists and transforms ad performance measurement, we encourage you to apply. Please prepare to share examples of a product or feature you built end to end, an AI feature you shipped with accuracy metrics, and a time when your evaluation caught a confident error and how you addressed it.