Amy AI
Led frontend engineering for a privacy-first AI companion, turning complex backend intelligence into a calm, safe, and intuitive teen-focused product experience.
For Amy AI, I owned the frontend experience across both the core product and waitlist platform, with a strong focus on emotional safety, clarity, and trust for teen users. I integrated the Next.js frontend deeply with FastAPI services, handling authentication, sessions, schema-safe request flows, loading/error states, and live user feedback while keeping the UI simple and reassuring.
Project At A Glance
Timeline
Jan 2026 – Present
Industry
AI · SaaS · Privacy
Contribution
UI/UX design, frontend system design, backend API integration, onboarding/chat experience, and smooth interaction delivery
Collaboration
Project Visual
Amy AI
Summary
What this project demanded.
For Amy AI, I owned the frontend experience across both the core product and waitlist platform, with a strong focus on emotional safety, clarity, and trust for teen users. I integrated the Next.js frontend deeply with FastAPI services, handling authentication, sessions, schema-safe request flows, loading/error states, and live user feedback while keeping the UI simple and reassuring.
Role
Frontend Engineer
Year
2026
Problem Space
Balancing sensitive mental-health UX with robust production integration: the interface needed to feel calm and non-overwhelming while still supporting real API workflows, anti-bot protections, security hardening, and future-ready streaming/chat behaviors.
I had to design for a sophisticated backend architecture that included an anti-sycophancy pipeline, LLM routing, streaming and non-streaming chat endpoints, Redis-backed state, PostgreSQL persistence, and future memory infrastructure. The UX challenge was making all of that complexity disappear for teen users.
Capability 01
Next.js
Capability 02
FastAPI
Capability 03
OpenRouter
Capability 04
Docker
Context
Amy was not a typical chatbot project. Its product promise depended on behavior.
The system architecture already had a strong technical backbone: FastAPI services, OpenRouter-based multi-model routing, an anti-sycophancy decision pipeline, Redis for state and rate limits, PostgreSQL for structured records, and Qdrant planned for memory. My work was to turn that into an interface that felt coherent, human, and teen-appropriate.
01
User needs
- A companion experience that feels validating without pretending to agree with everything
- A chat UI that stays calm and understandable during sensitive moments
- Smooth transitions through onboarding, login, and message-based flows without technical friction
02
System realities
- An anti-sycophancy classifier and challenge injector determine how the product should respond
- Routing spans GPT-4o-mini, DeepSeek-V3, and Claude safety checks behind OpenRouter
- The frontend must stay readable while sitting on top of health checks, API auth, retries, and stateful backend behavior
Signal
4
Core services exposed through the product architecture: FastAPI, PostgreSQL, Redis, and Qdrant
Signal
51
Backend tests already existed, so the UI had to respect a serious engineering system rather than act like a prototype
Signal
1
Unified product language created across waitlist, onboarding, and chat instead of fragmented screens
The design problem was translating deep backend rigor into emotional clarity.
Insight
Amy's best UX decision was restraint. When the product's intelligence depends on challenging distortions, the interface itself has to feel grounded and trustworthy.
Core Interaction Shifts
The product decisions that changed how the experience felt.
Shift 01
Trust-Centered Product Flows
Designed clean, low-friction interfaces with thoughtful messaging and interaction patterns that reinforce safety without adding cognitive overload.
Shift 02
Backend-Aligned Frontend Integration
Connected chat/forms to live FastAPI endpoints with session handling, API auth, request/response mapping, validation-safe payload transforms, and resilient error states.
Shift 03
Security + UX Refinement
Implemented bot protection and edge-case handling, removed enumeration-prone waitlist exposures, resolved mobile UX issues, and improved component structure for long-term scalability.
Influence & Validation
What changed because of the work.
The product moved from a technically capable AI system to a polished, safe, and user-centered experience that feels intentional in every interaction.
Frontend architecture aligned with anti-sycophancy behavior goals and future memory/streaming features
Production-grade integration across waitlist and core flows, including validation, rate-limit, and bot-protection edge cases
Mobile responsiveness, accessibility, and interaction consistency improved across critical user journeys