One workspace. Your data. A lighter runtime.
Affordable alternative to Overleaf & Teal
RoleTect’s two-award desktop app brings job capture, AI-assisted tailoring, and LaTeX compilation into a local-first career vault. A more affordable web SaaS alternative to Overleaf and Teal is in development with a Next.js frontend and Rust/Axum backend.
Product architect · Md. Ramjan Miah · 2026—Present
- Reported desktop runtime
- ~50 MB
- Instead of a 3–5GB+ TeX distribution; cached packages are additional.
- Document storage & rendering
- Local-first
- Cloud model requests are optional; local Ollama is an alternative.
- Desktop awards so far
- 2×
- 2nd runner-up at SciBlitz 2.0; 1st runner-up at BrainChild AI 2.0.
BENEATH THE INTERFACE
Inside a local-first desktop.
Peel back the workspace to see the boundaries that make it work.
01 / 05Start at the surface
One place to shape the next chapter.
Vue 3 and Pinia bring captured roles, source documents, and rendered PDFs into one desktop workspace. The interface is the entry point to the native services underneath.
The desktop interface; the Next.js web SaaS is a separate work in development.
Choose a chapter to explore the architecture.
Reduce friction, without outsourcing trust.
A fractured authoring loop
Copy a job posting, prompt an LLM, move the output to a LaTeX compiler, then track the application elsewhere. Each handoff loses context between the vacancy, source document, and exported PDF.
A heavyweight starting point
Traditional TeX distributions and container images can take several gigabytes. A non-technical user should not need to manage a typesetting environment just to edit a résumé.
Convenience versus data custody
Career documents contain detailed personal information. Centralized storage is convenient, but introduces another organization into the document’s trust boundary.
From a job posting to a local PDF.
Four boundaries keep the workflow understandable: capture, inbox staging, model orchestration, and native compilation.
- 01
Capture
Browser extension
A visual DOM selector extracts job content and sanitizes the payload before sending it to the local ingress server.
- 02
Stage
Axum / Tokio → SQLite
An asynchronous local receiver stores raw jobs in an inbox. A 20-port discovery pool avoids a rigid, collision-prone socket.
- 03
Orchestrate
Vue 3 / Pinia → Tauri IPC → Rig
The interface requests structured job metadata and context-aware document tailoring through a provider-agnostic Rust layer.
- 04
Compile & recover
Tectonic / Rust workers
Native typesetting produces a local PDF. Error logs become available after compilation; failed renders preserve the source and log for an AI-assisted correction and retry.
The decisions behind the interface.
Embed the compiler, not an entire distribution
Tectonic runs in asynchronous Rust workers, separating document rendering from the UI. Packages and macros are cached on first use, rather than requiring a full TeX Live installation.
Trade-off / First compilation still needs network access for missing packages. Runtime size is not the total installed cache size, and compilation time depends on the document and host hardware.
Keep ingestion local and explicitly staged
The extension sends sanitized content to an Axum receiver. SQLite inbox staging separates raw captures from validated application records; Vue 3 and Pinia coordinate the interface through Tauri commands.
Trade-off / Local service discovery and payload validation remain important boundaries. A loopback endpoint is an architectural choice, not a blanket security guarantee.
Make the model a choice, not a dependency
Rig abstracts multi-provider requests for parsing and tailoring. The report describes commercial providers, custom endpoints, and local Ollama models, with credentials kept in an encrypted vault.
Trade-off / Cloud providers receive the context sent in a request. Fully on-device AI requires a local model; encryption at rest does not make external inference air-gapped.
A workspace, not a language-server-backed editor
RoleTect supports document authoring and compilation. It has no language server (LSP) or live editor diagnostics; LaTeX errors are reported in the compiler log only after a compile attempt.
Trade-off / Feedback is compile-driven rather than continuous. Users compile the document to inspect errors, then edit the source or request an AI-assisted correction.
Treat compiler errors as structured feedback
On failure, the workflow records the broken LaTeX and Tectonic log, sends that context for a proposed correction, then attempts compilation again. Successful output clears the recorded error state.
Trade-off / Recovery needs model credentials and is not guaranteed to succeed. The useful outcome is a verified render, not merely a plausible model response.
What runs locally. What needs configuration.
Local workspace
- LaTeX editing and local rendering after package caching
- Native Markdown authoring and preview
- Mermaid diagram authoring and high-resolution exports
- Browser capture into a local inbox
With credentials
- Structured job parsing and résumé tailoring with a configured model
- AI-assisted LaTeX repair using compiler feedback
- S3 snapshot backups with user-provided storage credentials
Web SaaS · In development
A more affordable alternative to Overleaf and Teal, built with Next.js + Rust + Axum. The two awards belong to the desktop app, not the web SaaS.
Roadmap — not presented as shipped- Web SaaS with a Next.js frontend and Rust/Axum backend — in development
- Private deployments for mobile access and link analytics
- Automated snapshot synchronization and off-site lifecycle policies
- First-use package hydration and simpler cross-platform installation
Based on the supplied June 2026 architecture report and résumé. Runtime figures are author-reported, not independently benchmarked; latency and privacy depend on configuration.