WorkServicesExperienceAboutContact
01 / Case StudyAI Dream Interpreter

DREAM VEIL

An AI dream journal with real-time interpretation, semantic pattern matching, and a zero-knowledge Blind Vault.

DREAM VEIL interface preview

Role

Full stack architecture, E2EE systems, RAG pipeline, AI orchestration

Timeline

2026 - Present

Stack

Next.js / FastAPI / pgvector / Gemini 3.5

Project Context

Built ToFeel Useful.

Dream Veil is an AI-powered dream journaling platform built around contextual interpretation and semantic pattern recognition. Users record dreams, receive streaming psychoanalysis from user-selectable Gemini 3.5/2.5 Flash models with configurable thinking depth and optional multimodal image analysis, and discover thematic links across their history all while optional E2EE ensures the server never permanently stores plaintext narratives or interpretations.

Architecture

Zero-Knowledge

Retrieval

Hybrid RAG

Real-Time

SSE Streaming

Technical System

Sharp stack. Clear product shape.

Next.jsTypeScriptReduxIndexedDBTailwind CSSWeb Crypto APIFastAPISQLAlchemy 2.0AlembicPostgreSQLpgvectorRedisSSEGemini 3.5 FlashGemini 2.5 Flashgemini-embedding-001Cross-Encoder RerankBM25 / RRFOAuth / JWTDockerAWS Lambda
01

Architected the Blind Vault a 4-phase zero-knowledge E2EE handshake across REST and SSE with a 5-phase resumable batch migration state machine for large libraries.

02

Built a 6-phase production RAG pipeline: hybrid pgvector + BM25 retrieval, reciprocal rank fusion, cross-encoder reranking, grounding guards, and tiered chronological context assembly.

03

Streamed multimodal AI interpretations via Server-Sent Events with BYOK Gemini keys, Redis-backed RAG context caching, and intelligent API key rotation with tiered TTL exhaustion handling.

Visual Archive

Screens

01

Next Step

Explore the live build.