Commit Graph

41 Commits

Author SHA1 Message Date
db6004fadc Implement support for segments with an optional categories key 2026-04-28 22:41:16 -05:00
177e45f1fd Removed historical prototype code 2026-04-28 22:35:32 -05:00
e71e5b8faf Replaced the deterministic GrammarOnlyValidator with an LLM-backed grammar_only_guard 2026-04-28 22:29:28 -05:00
bbbef37d9a Implemented optional concurrency for the LLM backend 2026-04-28 22:14:25 -05:00
f834ffad97 Implemented provider-neutral OpenAI-compatible LLM endpoint support 2026-04-28 21:45:38 -05:00
8cab965abf Implemented support for the seriatim JSON schema for input transcripts 2026-04-27 22:11:19 -05:00
b9eef0043b Updated the spoken-word module prompts to preserve intentional emphatic repetition 2026-04-25 14:46:19 -05:00
a255ff28b7 Added new deterministic validators to catch early failures 2026-04-25 14:14:43 -05:00
619c615d21 Implemented a custom validator for the glossary module that allows changes from one protected glossary term to another 2026-04-25 13:58:20 -05:00
bf05d79fc6 Added a validator to prevent changes that result in an empty or whitespace-only string 2026-04-25 13:42:18 -05:00
5944f13404 Narrowed the grammar module LLM prompt and added support for a <-> an replacements 2026-04-25 13:25:19 -05:00
4724cd16c8 Implemented the grammar LLM review module 2026-04-25 09:06:06 -05:00
52d29f7228 Implemented the spoken_word LLM review module 2026-04-25 08:12:36 -05:00
92c8c371a6 Implemented a --modules CLI flag to allow runtime selection of the modules to be run 2026-04-25 07:48:21 -05:00
2d1d21d314 Completed the MVP for the new Audita refactor 2026-04-24 13:17:36 -05:00
34b3c09e43 Replaced the stub LLM-based validator class with two real LLM-based validator implementations 2026-04-24 11:52:22 -05:00
bca2152971 Added an interface for validators, and implemented an initial deterministic validator to protect terms in the provided glossary 2026-04-24 11:08:32 -05:00
f39de37974 Moved the current implementation to src/audita_prototype, moved current test suite to tests/audita_prototype, and started a new, more modular application skeleton in src/audita 2026-04-24 10:17:38 -05:00
889b620f2e Update default max segment gap to 4.0 and adjust related tests 2026-04-23 20:50:12 -05:00
3277587e3a Improvements around reporting and work directory retention options 2026-04-23 12:08:44 -05:00
c4e2db75f1 Implemented a second-stage validator for grammar corrections that allows homophone-related changes 2026-04-23 11:25:35 -05:00
dee6ae4067 More bugfixes in the glossary protection guard 2026-04-23 10:51:47 -05:00
671188aaba Update README with new configuration options and defaults 2026-04-23 07:08:58 -05:00
10bec989ce Raise default LLM confidence thresholds and automatically retry segments with low-confidence corrections 2026-04-22 17:50:18 -05:00
d7788da1f4 Implemented a validation step to confirm that grammatical changes do not change substantive meaning 2026-04-22 15:58:05 -05:00
87c47517a0 More bugfixes in the glossary protection guard module 2026-04-22 12:17:52 -05:00
15cdf01f58 Bugfixes in the LLM retry loop 2026-04-22 11:36:24 -05:00
13b81aab2a Extend glossary term protection to the initial glossary LLM stage 2026-04-22 11:15:36 -05:00
183f8ba31d Bugfixes in the glossary protection guard: allow changes toward protected glossary terms, but deny changes away from protected terms 2026-04-22 11:04:35 -05:00
9afb5c5a4e Grammar corrections now run through a deterministic guard that blocks protected glossary names/aliases from being changed 2026-04-22 09:04:00 -05:00
e50124fa0d Updated default config values to improve reliability and readability 2026-04-22 08:47:22 -05:00
c1672af9b4 Implemented deterministic transcript normalization before the LLM stages 2026-04-22 07:22:40 -05:00
445329de81 Implemented a second LLM stage for grammatical review 2026-04-21 15:42:09 -05:00
ca01e46d77 Updated configuration to utilize segment ids provided in the input transcript 2026-04-21 15:01:31 -05:00
7e1a3f721a Added support for multiple LLM passes to correct the transcript based upon the provided glossary 2026-04-21 13:12:08 -05:00
4296e3576e Refactor to request matching/replacement substrings from LLMs, rather than requesting a complete replacement for the full original text 2026-04-21 12:33:15 -05:00
23532cade1 Simplified the json schema passed to the LLM, and implemented more forgiving error handling for LLM proposed corrections 2026-04-21 11:14:56 -05:00
8c80c942dd Updated the LLM prompt to enforce acoustic/phoenetic similarity requirements for proposed corrections 2026-04-21 10:46:00 -05:00
8f6c721f06 Added executable root launcher for audita 2026-04-21 10:19:39 -05:00
e10349302f Initial version of application 2026-04-21 10:00:01 -05:00
fd0f57274d Initial commit 2026-04-21 14:01:41 +00:00