Method
Speech recognition
ASR models and ensembling (Conformer, RNN-T, wav2vec, ROVER).
13 cases from 11 companies apply Speech recognition. 4 report results in production.
Typical tasks: Moderate content at scale, Reduce ML cost, Evaluate models, Catch fraud and abuse and 4 more. Companies: Avito, CFT, X5 Tech, Celsus, Algolia, Snap and 5 more. Often combined with: Data augmentation, Full-text search, Multimodal embeddings, Embedding similarity matching, Pseudo-labeling and 38 more.
A Speech Platform in Production![]() | ![]() | retail | 2026-05 |
| AI Assistant for Sales | classifieds | 2026-05 | |
A Synthetic Training Pipeline for Medical LLMs![]() | ![]() | healthtech | 2026-05 |
| Which episode was that? Making your favorite TV show searchable | media-streaming | 2025-07 | |
| Complex and Urgent: Cutting Labeling Costs Without Losing QualityMA | MA | telecom | 2025-05 |
The onnx-asr Library: Speech Recognition in Python with ONNX Models (Without PyTorch)![]() | ![]() | general | 2025-05 |
| Snap Ads Understanding: From Pixels to Words | adtech | 2025-04 | |
From Platonic Dialogues to Strong AI, or a Few Words on Multimodal RAG for LLMs![]() | ![]() | general | 2024-05 |
| LLM Practice Without ChristofariCF | CF | general | 2024-05 |
| Launching Automated Video Moderation | classifieds | 2024-05 | |
Clean Labels for ML![]() | ![]() | search-engine | 2023-05 |
| SpectroBrain: Detecting Phone Spam with Semi-Supervised Learning | marketplace | 2023-05 | |
| Improving speech recognition without labels, sign-up or SMSCF | CF | general | 2022-10 |




