Initializing System StateSUB-100MS EDGE SLA // INGRESS ROUTING
Back to Case Studies
Tier 2 — Production Project Spec
AI SaaS / 2025
Quill — AI-Powered PDF Chat
AI-powered PDF analysis platform enabling semantic search and conversational queries across document collections.
RoleProduct Engineer
Duration2 months
“Why this system existed: AI-powered PDF analysis platform enabling semantic search and conversational queries across document collections. We engineered explicit boundaries to prevent race conditions and ensure predictable failover under production traffic.”
Executive Summary
Production SaaS combining Pinecone's vector database with LangChain's RAG pipeline for document semantic understanding. Features real-time streaming responses with Gemini API and freemium monetization with Stripe webhooks.
02 — Business Context
Why This Exists
Professionals spend hours reading and searching through PDF documents. Existing PDF tools only offer keyword search, not semantic understanding of document content.
03 — Problem Statement
Technical Formulation
Create a platform where users can upload PDF documents and ask natural language questions, receiving answers with direct citations to source content.
System Specification
- 01Upload and process PDF documents of varying sizes
- 02Semantic search across document collections using vector embeddings
- 03Real-time streaming AI responses for responsive UX
- 04Freemium pricing model with usage limits
- 05Direct citations linking answers to source pages
06 — Architecture
System Design & Data Flow
Next.js frontend with API routes. Pinecone vector database for embeddings storage. LangChain for RAG pipeline. Gemini API for LLM. Supabase for file storage. Stripe for subscription management.
System Architecture & Data Flow Invariant
ASCII TOPOLOGY+-------------------------------------------------------------------------+ | SYSTEM PIPELINE INVARIANT | | | | [Client Ingress] ---> [Edge Router & Auth] ---> [Transactional Store] | | | | | | v v | | [Async Workers] <------- [Event Queue / Broker] | +-------------------------------------------------------------------------+
08 — Engineering Decisions
Architectural Choices
- Decision #1Selected Pinecone for vector storage — better suited for document embeddings that don't need relational queries
- Decision #2Used Supabase Storage instead of S3 — simpler integration with existing Next.js project
- Decision #3Implemented streaming responses for better UX despite added complexity
09 — Trade-Offs
Managed Compromises
- Trade-Off #1Pinecone vs pgvector — Pinecone is managed (easier) but more expensive at scale
- Trade-Off #2Streaming responses increase frontend complexity but significantly improve perceived performance
Production Impact
PDF Processing1s/page
Search Latency<50ms
Response Start<500ms
- Context-aware AI responses using LangChain and Gemini API
- Real-time text streaming for seamless chat experience
- Freemium model with automated Stripe subscription management
- Secure file storage and retrieval with Supabase integration
11 — Lessons Learned
Technical Post-Mortem
- • PDF parsing quality varies significantly between libraries — PDF parsing is the primary bottleneck for answer accuracy
- • Chunk overlap size significantly impacts retrieval quality for dense technical documents
12 — Future Work
Scalability Roadmap
- • Continuous performance profiling and edge latency optimization.
- • Expanding API integration endpoints and developer documentation.
Field Notes & Architectural Decisions
Related Engineering Notes
Continue Exploring
Interconnected Engineering Knowledge
This project demonstrates
- Event-driven SaaS architectures
- Sub-second edge ingestion
- Atomic database usage metering
- Zero-config webhook security
Related projects
Technologies used
Next.jsLangChainGemini AIPineconeSupabaseStripe
Building something with similar constraints?
Let's evaluate your technical requirements and outline an execution plan.