Source Library
Source Library is a web platform that makes rare Hermetic and esoteric texts accessible through digitization, AI-assisted processing, and modern web technologies. The project focuses on preserving cultural heritage while enabling new forms of research, interpretation, and discovery.
Key Contributions
Full-Stack Ownership & Collaboration · System Evolution · Delivery
-
Initially contributed as a backend engineer alongside a frontend developer.
-
Took full ownership of frontend development after the frontend developer's contract ended, transitioning into a full-stack role.
-
Worked closely with CEO to continue delivery without disrupting timelines or system stability.
-
Ensured continuity of development while expanding system capabilities.
Multi-Tenant System Architecture · Scalability · Isolation
-
Designed and implemented a multi-tenant architecture to support multiple organizations/clients on a single platform.
-
Applied the Bulkhead design pattern to isolate tenants across different AWS and MongoDB regions and connections.
-
Implemented backend middleware to resolve tenant context from a request header (
tenant slug) and route requests accordingly. -
Added an LRU cache layer to optimize tenant resolution lookups and reduce database overhead.
Authentication & Authorization · Security · RBAC
-
Implemented secure authentication using Argon2 with salt and pepper for password hashing.
-
Designed a JWT-based authentication system with access and refresh tokens.
-
Refresh tokens:
- Stored as HTTP-only cookies
- Long TTL (20 days)
- Revoked on logout and tracked in a dedicated revocation store
-
Access tokens:
- Short TTL (20 minutes)
- Managed in-memory on the frontend client
-
Implemented role-based access control (RBAC) across backend APIs and frontend UI.
-
Defined hierarchical roles: super admin, admin, editor, user, guest.
-
Enforced authorization on backend endpoints even if frontend checks are bypassed.
Frontend Infrastructure · Client Enforcement · Consistency
-
Implemented Axios interceptors to automatically attach:
- Tenant slug headers
- Authorization headers when access tokens are available
-
Controlled UI rendering and route access based on user roles and permissions.
-
Ensured frontend behavior remained consistent with backend authorization rules.
AI-Assisted Document Processing Pipeline · OCR · Language Models
-
Integrated multiple AI providers including Gemini, OpenAI, and Mistral.
-
Built an end-to-end processing flow where users upload scanned PDFs.
-
Used AI to:
- Split double-page book scans into individual pages
- Perform OCR on scanned pages
- Translate extracted text
-
Implemented tooling to:
- Re-split incorrectly processed pages
- Reorder pages to restore correct reading sequence