TexFolio

The Problem

Most resume builders generate messy HTML-to-PDF exports that fail ATS parsers. LaTeX produces perfect typography but has a steep learning curve. Existing tools lack AI feedback, team collaboration, and enterprise features like RBAC and GDPR compliance.

The Solution

Created a "No-Code LaTeX" SaaS with a Hono backend that injects user data into .tex templates via Mustache and compiles them using a Dockerized pdflatex environment. Integrated a LangGraph multi-agent pipeline (4 stages: Content, Impact, Format, ATS) for intelligent resume coaching. Added Organizations with full RBAC hierarchy, BullMQ async PDF generation with progress tracking, distributed Redis rate limiting, and GDPR-compliant data export/deletion.

Key Challenges

• 1Building a LangGraph multi-agent pipeline with circuit breaker failover (NVIDIA NIM → Groq) that scores resumes across 4 dimensions and provides actionable feedback.
• 2Implementing Organizations with RBAC — ownership transfer, org-scoped resume visibility, branding overrides (locked templates, company fonts, colors) applied during PDF generation.
• 3Designing BullMQ async PDF generation with progress tracking (10% → 30% → 100%), automatic retries with exponential backoff, and rate-limited worker concurrency.
• 4Sandboxing LaTeX compilation using spawn (not exec) to prevent shell injection while supporting custom templates.

Tech Stack