Senior AI Backend Engineer (Psychology Stack)

**About PixaSocial Ai** PixaSocial Ai (https://pixasocial.com) creates viral social content powered by proprietary psychology and psychographic models. It generates posts, visuals, and campaigns, then publishes across Instagram, X, LinkedIn, and 12\+ other channels. Product surfaces include Brand Influencers, Product UGC Studio, Reel Factory, Media Studio, image/video/voice generation, and agentic campaign operators with persona export. Sign in at https://app.pixasocial.com. Legal entity: Pixasocial LLC, 1209 Mountain Road PL NE Ste N, Albuquerque, NM 87110, USA. Careers listing: https://pixasocial.com/careers (currently LinkedIn\-only; apply for this role by email as below). About the role This is the backend job closest to PixaSocial’s edge: the psychology and psychographic layer that sits in front of image, video, copy, and campaign generation. You will design and maintain the services that turn audience briefs into scored personas, strategy signals, and prompts that the rest of the product actually uses—not a research demo, not a one\-off notebook. We are early\-stage. You will own services end to end, make calls about APIs and data contracts, and sit in the path between “psych model” and “post that ships to Instagram, X, LinkedIn, and 12\+ other channels.” You will use the product. If you have not built an audience, generated a campaign, and tried to publish, you will not interview. What you’ll do \- Design, ship, and operate core backend services that power proprietary psychology and psychographic scoring—audience models, persona objects, strategy signals, and the APIs that content generation calls. \- Define stable contracts so Brand Influencers, Product UGC Studio, Reel Factory, Media Studio, and the campaign operators all consume the same psych layer instead of forking logic. \- Build and harden scoring / ranking pipelines: inputs from a brief (goal, audience, channel), outputs that route into copy, image, and video generation without leaking proprietary methods into logs or client payloads. \- Put evals around psychographic quality, not just latency: regression tests when a scoring change would silently worsen captions, persona export, or campaign operators. \- Work with queues, persistence, and APIs so scoring is async where it should be, cached where it can be, and correct when a user hits Publish. \- Instrument services so we can see failure modes (timeouts, empty personas, bad channel mapping) before a marketer does. \- Pair with platform and applied\-AI engineers on inference cost, batching, and what the psych stack should \*not\* call a frontier model for. \- Use PixaSocial weekly as a customer would. File what is broken in the psychology → content → publish path, then fix the backend slice you own. What we’re looking for \- Several years building production backend services (mid\-senior): APIs, background jobs, data stores, not only notebooks. \- You have shipped systems that score, rank, or personalize—recommenda