Qwen3.6-35B-A3B automates document workflow from evaluation to PDF
One Reddit user's local LLM handles database queries, emails, Google Docs, and PDFs.
A Reddit user has detailed a practical, multi-step workflow powered entirely by local LLMs, specifically Qwen3.6-35B-A3B (a 35B-parameter model with 3B active parameters via mixture-of-experts). The workflow starts with embedding models that give the AI a semantic search protocol for persistent memory, making recall feel seamless. On a weekly schedule, the LLM evaluates a database against user-defined criteria, then emails a summary of matching items.
The user responds via email with selections, and Qwen takes those choices to query sources and a knowledge base, generating a draft document in Google Docs. The user edits the doc, leaves comments, and Qwen incorporates feedback. Once iteration is done, an email triggers conversion to PDF. The entire loop—database scanning, email back-and-forth, document editing, and export—runs locally, without any cloud API calls. The user reports the model "knocks down every task" and plans to scale complexity. This real-world example counters skepticism that local LLMs can handle structured, multi-step automation for knowledge work.
- Uses embedding models for persistent memory with semantic search, enabling seamless recall in a local AI harness.
- Qwen3.6-35B-A3B runs a weekly automated database evaluation, emails results, and iterates on documents via Google Docs integration.
- The workflow concludes with LLM-generated PDFs—all done locally, proving local models can replace cloud-based automation pipelines.
Why It Matters
Local LLMs like Qwen can automate complex, multi-step office workflows without cloud costs or privacy trade-offs.