 | Hi there. Not native English speaker. Not AI edited, so bear with me. 15+ years as social researcher for public bodies (currently unemployed). A lot of Policy Brief, reports and similar docs for higher ups in Government and Public Administration. Wanted to try qwen3.6-35b-a3b in Hermes Agent to make deep research and write well built reports, but the included skill feels lacking. However, for the first time with the Qwen model, I felt it was possible to achieve something similar to Perplexity. And after some work and five hours of the machine humming in a corner, it produced something quite acceptable. No excellent, but good enough to start with. Six loops in total over the same document (21 pages), from draft to diagnose problems and fixing it, making charts and inserting it. Almost autonomously. I think that it can go complete autopilot in the future, with very precise prompts. Also: More than five hours non stop. 28 tokens per seconds. Slow. (12th Gen Intel Core, 32 Gb RAM, RTX 4060, LinuxMint) To anyone curious, the git repo with all the skills, prompts, meta-prompts, python scripts and all intermediate artifacts, including the final report made by the agent (on the current state of AI in Europe, md, docx and pdf format). The readme and folder organization was made by the same AI agent (too busy / lazy to care about) However I think that can be interesting to anyone in the public research business, to use it as a first step. I recommend to use an AI to navigate the documents and folders. [link] [comments] |