AI Product Manager — Thrifle AI

About Thrifle

Thrifle.com is the most intelligent shopping platform in the United States. We're building the country's largest free database of retailer return policies (150+ retailers graded A+ to D), real-time CPSC and FDA product recall cross-checking, military and student discount verification, live Bureau of Labor Statistics Consumer Price Index data, EIA gas prices, and Federal Reserve borrowing rates — all surfaced inside the shopping flow so consumers can answer "Is this actually a good deal for me?" before they buy.

We are not a coupon aggregator. We are not a price comparison engine. We are an AI-first shopping intelligence platform with proprietary data, a custom multi-phase AI content pipeline (Anthropic Claude Sonnet 4.6 + Haiku 4.5), and live data integrations across federal economic datasets — building the most comprehensive consumer shopping intelligence layer on the internet.

We're a lean, founder-led team shipping fast.

The Role

We're hiring an AI Product Manager to own Thrifle AI — the proprietary multi-phase AI content + intelligence pipeline that powers every deal page, blog post, FAQ generation, and recall match on the platform.

Thrifle AI today runs production Anthropic Claude pipelines for: deal description enrichment, multi-phase research-grounded blog generation (research → brief → section-by-section), retailer FAQ generation (1,500+ FAQs shipped), recall semantic matching, retailer about-content generation, and structured-data extraction. You'll own the roadmap, the prompts, the evals, and the user-facing AI surfaces.

What You'll Own

• Thrifle AI roadmap: the next 6–12 months of AI features, from prompt engineering and eval design through to user-facing AI surfaces

• Prompt + eval infrastructure: testing harnesses, hallucination detection, prompt versioning, model-version rollouts

• Cost + quality optimization: Sonnet vs Haiku routing, prompt caching, ephemeral context windows, token budgeting at scale

• Voice + style enforcement: we have a 29-pattern AI-tells stripper, em-dash budgeting, banned-phrase enforcement — extend and improve

• New AI features: "Should I buy this now?" verdict engine, real-time price-anomaly detection, semantic deal clustering, multimodal product matching

Tech / Tools You'll Work With

• Anthropic Claude SDK (Sonnet 4.6, Haiku 4.5), prompt caching, batch API

• Node.js + Express backend, Mongoose, Next.js front-end

• Vector stores, embeddings, structured JSON outputs

• Real-time data: CPSC, FDA openFDA, BLS API, FRED CSV

What We Want

• 3+ years as a PM or technical PM, with hands-on LLM application experience

• You can read and write code (Python/JS) — won't write the production code, but you'll prototype prompts, write eval scripts, and analyze cost/quality dashboards

• Strong instinct for "what's worth shipping" vs "what's a demo" in AI

• Bonus: prior experience in retail / e-commerce / shopping intelligence

Apply

Email [email protected] with: a 1-paragraph note on an AI product you've shipped, a link to a prompt or eval you're proud of, and the most interesting LLM problem you've worked on.