Academic/Technical Writer Needed: Convert ML Research Report into Formal Conference-Style Paper

We have a completed research project on audio-visual saliency prediction for 360° video — a multimodal transformer model called AV-SalViT that predicts where VR/AR users will look, using both visual and spatial audio cues. The technical work is done: architecture, training, and results all exist, and the project is currently written up as an informal report (around 14 pages, with a table of contents, a section-by-section narrative, figures, and a reference list). We're looking for an experienced technical or academic writer who can turn this into a polished, formal paper suitable for submission to a workshop or conference, and who is also comfortable strengthening the technical content itself where it would make the paper more compelling, not just reformatting the prose. We'll provide the full current draft as a PDF, covering the introduction and motivation, the visual saliency pipeline, the audio saliency pipeline, the complete model architecture with diagrams, training methodology, evaluation metrics (NSS and CC), results, and a 14-item reference list. All existing figures and diagrams can be reused as-is or redesigned if that improves clarity. The core of the work is restructuring the report into a standard academic paper format — Abstract, Introduction, Related Work, Method, Experiments and Results, Discussion, and Conclusion — following the conventions of whatever venue we target. That includes writing a proper Abstract and a genuine Related Work section that synthesizes the existing references (covering audio-visual saliency, omnidirectional video saliency, and spherical vision transformers) and clearly positions our contribution against prior work, something the current draft doesn't really attempt. Beyond restructuring, we want the writer to push on the substance: tightening the Method section so the architecture, fusion strategy, and probabilistic decoding are presented with real academic precision and consistent notation, sharpening the framing of our contributions, strengthening the experimental narrative and discussion of results, and flagging or fixing any places where the technical reasoning, claims, or evaluation could be made more rigorous or convincing. If something in the current draft is thin, hand-wavy, or could be argued better, we want that improved rather than just rephrased — feel free to suggest additional analysis, reframe weak claims, or recommend what's missing for a stronger paper, and we'll work with you on follow-up details as needed. A short Limitations and Future Work section should also be added, which is standard for formal papers and currently absent. The final deliverable should be full LaTeX source (.tex and .bib) plus a compiled PDF of the paper, formatted in a standard template — IEEE, ACM, or NeurIPS style, to be confirmed with you once we settle on a target venue — typically landing in the 6 to 10 page range excluding references, with citations cleaned up into proper BibTeX matching that venue's style. One round of revisions based on our feedback is included in the project. Ideal candidates have a strong academic or technical writing background, ideally with published papers in machine learning, computer vision, or HCI venues, are comfortable reading and meaningfully engaging with ML architecture descriptions (transformers, CNNs, multimodal fusion) well enough to improve how they're argued and not just transcribed, and are proficient in LaTeX using IEEE, ACM, or NeurIPS templates. Thank you!!