v1.1.0

What's Changed

• feat: multimodal support by @JamesMcMinn in #175
• 156 investigate multimodal analysis of short form videos by gemini spike by @dearden in #168
• Repair LLM output using json_repair by @andylolz in #178
• Update README.md regarding new users etc. by @dcorney in #180
• chore: adjust timeout to 2 hours by @JamesMcMinn in #182
• feat: updated gemini 1.5 to 2.0 by @dearden in #184

Full Changelog: v1.0.0...v1.1.0

The One Where We Do YouTube

What's Changed

• feat: change database schema by @ff-dh in #82
• refactor: replace react with a flask-only version by @andylolz in #78
• 70 change main inference model to multi label version by @dcorney in #76
• fix: more accurate chunk start and end offsets by @andylolz in #87
• 73-write-an-evaluation-script-using-promptfoo by @dearden in #83
• 95 Ask for a direct quote as well as a paraphrase in multi label prompt by @dearden in #96
• dev -> main. Schema changes; switch to flask-only; multi-label model by @andylolz in #84
• feat: COPY data in dockerfile to have training claims by @ff-dh in #97
• fix: store the sentence text, not the chunk by @andylolz in #98
• feat: refine claim offsets by @andylolz in #102
• fix: limit history to 20 items by @andylolz in #109
• feat: show transcript alongside claims by @andylolz in #110
• fix: update captions regex by @andylolz in #115
• fix: pass raw transcript text to LLM by @andylolz in #108
• 116 improve json parsing of generated output by @dearden in #117
• Evaluation assertions by @dearden in #112
• 81 evaluation write function to run evaluation code during promptfoo evaluation by @dearden in #94
• Evaluation test for checking if the quote appears in the chunk by @dearden in #99
• Add a test for checking the format of the LLM output by @dearden in #105
• feat: user registration routes by @ff-dh in #121
• fix: use user_id in runs & rename disable route by @ff-dh in #123
• chore: update readme for new users handling by @ff-dh in #128
• 131 remove evaluationpy by @dearden in #132
• fix: only call google.auth.default() from inside a function by @andylolz in #134
• fix: remove text from labels by @andylolz in #126
• fix: check end_idx before using it by @andylolz in #129
• 91 improve generation of labels for in context learning by @c-j-johnston in #127
• fix: add claim summaries again for app by @c-j-johnston in #135
• fix: add missing data file by @c-j-johnston in #137
• feat: parse output jsons using existing function by @c-j-johnston in #139
• feat: don't limit number of claims per chunk by @c-j-johnston in #142
• 140 add format checker by @c-j-johnston in #144
• refactor: assume we have exactly one candidate by @andylolz in #148
• feat: only show current user’s video history by @andylolz in #150
• feat: add summary to LLM output in evaluation by @c-j-johnston in #155
• fix: check dataframe columns by @c-j-johnston in #158
• 159 make package self contained by moving prompt data into src folder by @dcorney in #160
• feat: add notice to home page by @scoltman in #163
• Ensure example output is valid JSON by @andylolz in #172

New Contributors

• @scoltman made their first contribution in #163

Full Changelog: 0.0.1...v1.0.0

0.0.1

Initial Release

• database create script/module by @ff-dh in #10
• Rename FOREIGN KEY id -> video_id by @ff-dh in #13
• 3 enhance training data by @c-j-johnston in #15
• 4 make evaluation robust by @dcorney in #5
• 8 update list of search keywords and phrases to get a wider variety of health misinformation by @dearden in #19
• feat: merge backend to main by @ff-dh in #28
• feat: merge Frontend to main by @ff-dh in #27
• feat: ansible playbooks for deployment by @ff-dh in #30
• Copy changes by @andylolz in #32
• Remove sqlite database from source control by @andylolz in #46
• feature: use absolute imports with package modules by @ff-dh in #48
• build: install node dependencies with npm ci by @andylolz in #47
• feat: read baseUrl from an environment variable by @andylolz in #43
• docs: add ‘getting started’ instructions to README by @andylolz in #45
• refactor: remove tools.db by @andylolz in #52
• 24 use fact checker annotated data for in context learning by @dcorney in #50
• 58 add factchecker persona to prompts by @c-j-johnston in #59
• 56 chunks of youtube transcripts should include their timestamp offsets by @dcorney in #60
• feat: auto-deploy from GitHub Actions by @andylolz in #55
• fix: increase timeout from 30 seconds (default) to 30 minutes by @JamesMcMinn in #57
• refactor: rationalise into a single version by @andylolz in #63
• fix: unescape HTML entities in titles by @andylolz in #65
• feat: add MODEL_SHA to environment by @andylolz in #75
• 38 investigate genai evaluation tools by @dearden in #72
• 56 (take 2) chunks of youtube transcripts should include their timestamp offsets by @dcorney in #68

Full Changelog: https://github.com/FullFact/health-misinfo-shared/commits/0.0.1