YouTube IP V6

YouTube IP V6 is the current presentation-ready, Streamlit-deployable release. It carries forward the V5 product surface—seven sidebar destinations, public-only Channel Insights, bundled CSV benchmarking, and live YouTube / Gemini / OpenAI workflows—while refreshing the Streamlit entrypoint (streamlit_app.py → dashboard/app.py), built-in multipage navigation (st.navigation), dependency hardening (for example NumPy, PyArrow, and Google API client stacks on Windows and Cloud), and UI polish.

Live app (V6):

ip-youtube-creator-insight-2026.streamlit.app

Earlier V5 deployment (historical): youtube-ip-v5.streamlit.app

Quick jump:

If you want to understand...	Start here
current runtime and page flows	Architecture
deployment targets, secrets, and version mapping	Deployment And Versions
the full project story from V1 through V5 and into V6	Project Brief
how to explain Outlier Finder in a demo or presentation	Outlier Finder Presentation Notes
a GCP-friendly env template	`.env.gcp.example`

At A Glance

Metric	Current Count / State
Deployed versions documented here	`6` (`V1`–`V6`)
Live Streamlit app links (including historical)	`6`
Current V6 sidebar destinations	`7`
Current primary data paths	`2`
Current Channel Insights topic modes	`2`
Current live provider families	`3` (`YouTube`, `Gemini`, `OpenAI`)
Optional model-artifact path	`1` (`BERTopic` beta)

Current V6 In One Minute

V6 keeps the same two core data paths as V5: bundled GitHub CSVs for benchmarking and live API pulls for channel, outlier, tool, and thumbnail workflows.
Channel Insights remains public-only with two topic modes: default heuristics and optional BERTopic beta.
Thumbnails, Ytuber, and Tools continue the creative and AI-assisted workflows from V5.
V6 focuses on reliable deployment: explicit run() entry from streamlit_app.py, Streamlit’s st.navigation / pg.run() router, clearer dependency pins (pyarrow, google-api-python-client, google-auth-httplib2), and layout/CSS fixes for the global hero header.
V5 already removed the V4 Assistant and Google OAuth owner overlays; V6 does not bring them back.

Why This Project Exists

The original goal of the project was simple: help small-to-mid-sized YouTube creators make better content decisions with better intelligence than YouTube Studio alone provides. The early versions focused on public metadata, cross-channel benchmarking, semantic topic modeling, and AI-assisted recommendations so the team could answer questions like:

What content themes actually perform across comparable channels?
Which topics, formats, and publishing patterns correlate with stronger performance?
How can a creator move from raw channel data to a usable action plan for titles, thumbnails, and next videos?

That core question stayed constant across every version. What changed was the way the app packaged the answer.

Version Lineage (V1–V6)

Short history before the deeper tables below.

flowchart LR
    V1["V1<br/>analytics prototype"] --> V2["V2<br/>creator-suite expansion"]
    V2 --> V3["V3<br/>multi-page productization"]
    V3 --> V4["V4<br/>deepest intelligence layer"]
    V4 --> V5["V5<br/>cleanup + presentation-ready shell"]
    V5 --> V6["V6<br/>deploy + router refresh"]

    V1 --- A1["Public metadata, BERTopic framing, recommendation concept"]
    V2 --- A2["Ytuber suite, workflow expansion, AI generation, deploy practicality"]
    V3 --- A3["Clear page architecture, stronger runtime docs, outlier workflow"]
    V4 --- A4["Channel Insights, Assistant, Google OAuth, owner analytics, BERTopic beta"]
    V5 --- A5["Assistant removed, OAuth removed, public-only insights, retained AI suite"]
    V6 --- A6["st.navigation multipage, entry run loop, deps + header UX for Cloud/local"]

Deployed Version History

Version	Live App	Main Goal	Headline Additions	Major Simplifications / Later Changes	Status
`V1`	youtube-stats-ip.streamlit.app	prove the analytics + recommendation concept	public YouTube analytics framing, BERTopic modeling direction, initial Streamlit dashboard, thumbnail dashboard	later versions expanded beyond the original compact dashboard	historical prototype
`V2`	youtube-stats-ip-v2.streamlit.app	expand into a creator operating system	creator workflow framing, richer app shell, advanced Ytuber suite, deploy guidance	some modules were later split, simplified, or removed for maintainability	historical expansion
`V3`	youtube-ip-v3.streamlit.app	turn the project into a clearer product	five-page product shell, strong runtime architecture, outlier workflow, better repo map	later versions added deeper insights and then simplified again	historical productization
`V4`	youtube-ip-v4.streamlit.app	deepen intelligence and tracked-channel analysis	Channel Insights, sidebar Assistant, Google OAuth owner analytics, optional BERTopic beta	V5 removed the Assistant and Google OAuth to reduce deployment complexity	historical deep-intelligence release
`V5`	youtube-ip-v5.streamlit.app	keep the best parts and document the journey	public-only Channel Insights, retained AI suite pages, Thumbnails rename, consolidated docs, optional BERTopic beta	lighter than V4, easier to reason about, presentation-ready documentation	historical release
`V6`	ip-youtube-creator-insight-2026.streamlit.app	ship a dependable Cloud + local runtime on top of V5	Streamlit `st.navigation` router, `streamlit_app` → `run()` entry, explicit PyArrow / Google client deps, sidebar + hero layout fixes	product scope matches V5; focus is operational reliability	current release

What Changed Across Versions

Area	V1	V2	V3	V4	V5	V6
Core problem	creator strategy from public data	same	same	same	same	same
Bundled dataset benchmarking	present	present	strong	present	present	present
Creator workspace	early	expanded heavily	strong	present	present	present
Outlier research	emerging	present in suite	strong standalone page	strong	strong	strong
Thumbnail generation	present	expanded	present	present	present	present
Channel Insights tracked snapshots	absent	absent	absent	added	retained	retained
Sidebar Assistant	absent	absent	absent	added	removed	removed
Google OAuth / owner analytics	absent	absent	absent	added	removed	removed
Optional BERTopic model path	conceptual	conceptual	documented stack direction	added to runtime	retained	retained
Presentation-quality consolidated docs	basic	practical	stronger	strong	strongest	strongest
Streamlit multipage router (`st.navigation`)	—	—	—	—	custom / evolving	built-in

What Survived Into V6

Public-data-first creator intelligence stayed central from V1 onward.
Bundled CSV benchmarking survived as Channel Analysis.
Live public-channel analysis survived and matured into Channel Insights.
AI-assisted creative generation survived through Thumbnails and Ytuber.
Outlier research survived and remains a distinct workflow.
Optional BERTopic topic modeling survived, but as a guarded beta path rather than a required dependency.
Streamlit deployment remained the delivery surface for every released version.

What We Tried And Later Removed Or Simplified

Capability	Highest-Version Form	Why It Was Valuable	Why It Was Reduced Or Removed In V5 / Unchanged In V6
Sidebar `Assistant`	V4	gave global help and retrieval-driven guidance across pages	added surface area, more maintenance, and more cognitive load than the lighter shell needed; still absent in V6
Google OAuth + owner analytics overlay	V4	enabled private metrics such as impressions, CTR, watch time, and retention	raised deploy complexity, secrets burden, and public-vs-owner workflow branching; still absent in V6
Heavier mixed recommendation UI	V3 / V4	combined strategy guidance with creative tooling	overlapped with `Channel Analysis` and `Channel Insights`; V5 narrows page 3 to `Thumbnails` (unchanged in V6)
Broader “everything page” behavior in earlier suites	V2	useful during exploration and experimentation	harder to document and reason about than clearly separated workflows
Always-growing feature surface	V2-V4	helped the team test many ideas quickly	eventually created duplication, deployment weight, and documentation drift

Current V6 Product Surface

The current V6 sidebar order is:

Channel Analysis
Channel Insights
Thumbnails
Outlier Finder
Ytuber
Tools
Deployment

This is the high-level page map. The detailed runtime handoffs for each page live in Architecture.

Page	What Problem It Solves	Main Inputs	Main Outputs	Runtime Type
`Channel Analysis`	benchmark bundled datasets and compare portfolio-level performance	committed CSV data in `data/youtube api data/`	KPI cards, trend charts, channel/video rankings	dataset-backed
`Channel Insights`	analyze one tracked public channel over time	live YouTube Data API pulls, snapshot history, optional BERTopic	topic trends, format patterns, outliers, next-topic ideas	mixed
`Thumbnails`	generate and export thumbnails without mixing broader strategy UI	Gemini/OpenAI image calls, public thumbnail URLs	generated concepts, preview cards, downloads	mixed
`Outlier Finder`	find breakout videos in a niche	live YouTube API scans and outlier scoring	scored outlier tables, breakout snapshots, AI research	mixed
`Ytuber`	run a creator-focused live workspace	live channel data, AI generation, creator tools	audits, planner outputs, AI Studio results	mixed
`Tools`	inspect and export public YouTube assets	YouTube metadata, transcripts, yt-dlp, ffmpeg	previews, transcript exports, audio/video/thumbnail downloads	API-backed
`Deployment`	show deployment/setup guidance inside the app	static in-app instructions	repo, branch, secrets, deployment notes	static

Current V6 Workflow Map

This section is about the live app today, not the historical versions.

Page	User Goal	Main Inputs	Main Services Used	Main Outputs	Runtime Type
`Channel Analysis`	compare bundled channel/video benchmarks	committed CSV files, filters, date ranges	pandas transforms, visualization helpers	KPI cards, monthly trends, top channels, top videos	dataset-backed
`Channel Insights`	analyze a tracked public channel over time	channel URL/handle/ID, optional beta topic mode, snapshot refresh actions	`public_channel_service`, `channel_insights_service`, `channel_snapshot_store`, `topic_model_runtime`, `model_artifact_service`	topic trends, format metrics, outliers, next-topic ideas, history	mixed
`Thumbnails`	generate new thumbnail concepts or export a public one	title/context/style prompts, provider/model choice, YouTube video URL or ID	`thumbnail_generator.py`, `thumbnail_hub_service.py`	generated images, preview cards, prepared downloads	mixed
`Outlier Finder`	surface overperforming videos in a niche	niche query, filters, optional AI research trigger	`outliers_finder.py`, `outlier_ai.py`, provider-key helpers	outlier cards, scored result tables, breakout charts, AI insight cards	mixed
`Ytuber`	open a live creator workspace for one channel	channel query, live refresh toggle, segmented workspace module selection	YouTube API loaders, keyword/title scoring helpers, thumbnail generator, outlier handoff logic	audit views, keyword tables, AI Studio outputs, planner and benchmark results	mixed
`Tools`	inspect public YouTube assets and prepare downloads	single URL, batch URLs, playlist URL, operation choice	`youtube_tools.py`, `transcript_service.py`, `yt-dlp`, `ffmpeg`	metadata previews, transcripts, audio/video/thumbnail artifacts, batch/playlist results	API-backed
`Deployment`	understand how to run and deploy the app	none; in-app reference content	app shell guidance in `dashboard/app.py`	deployment instructions, repo/branch/secrets notes	static

flowchart TD
    A["Bundled CSV data"] --> B["Channel Analysis"]
    A --> C["Historical benchmark context"]

    S["Streamlit secrets / env"] --> K["src/utils/api_keys.py"]
    K --> Y["YouTube Data API"]
    K --> G["Gemini / OpenAI"]

    Y --> D["Channel Insights"]
    Y --> E["Outlier Finder"]
    Y --> F["Ytuber"]
    Y --> H["Tools"]
    Y --> I["Thumbnails URL export"]

    G --> J["Thumbnails generation"]
    G --> L["Ytuber AI Studio"]
    G --> M["Outlier AI research"]

    D --> N["Snapshots + topic metrics + recommendations"]
    E --> O["Outlier results + breakout charts"]
    F --> P["Creator workspace modules"]
    H --> Q["Prepared public asset downloads"]
    J --> R["Generated thumbnail images"]
    I --> T["Prepared thumbnail download"]

In practice, V6 works as three layers:

Data: bundled CSV benchmarking plus live provider/API calls
Service: normalization, scoring, topic assignment, artifact prep
UI: Streamlit cards, charts, tabs, downloads, and guided workflows

For the full section-by-section mechanics, see Architecture.

Current Interactive Surfaces

These are the main interactive surfaces a user actually navigates once a page is open.

Page	Surface Type	Count	Current Surfaces	What They Do
`Channel Analysis`	main analytics canvas	`1`	dataset filters + charts/tables	benchmark bundled CSV data
`Channel Insights`	tabs	`6`	`Overview`, `Topic Trends`, `Formats & Patterns`, `Outliers`, `Next Topics`, `History`	turn tracked public-channel snapshots into interpretable strategy signals
`Thumbnails`	tabs	`2`	`Generate`, `Download From URL`	create new thumbnails or export a public one
`Outlier Finder`	post-search sections	`4`	`Top Outliers In This Scan`, `Breakout Snapshot`, `AI Research`, `How This Works`	score breakout videos first, then interpret them
`Ytuber`	segmented modules	`8`	`AI Studio`, `Overview`, `Channel Audit`, `Keyword Intel`, `Outliers Finder`, `Title & SEO Lab`, `Competitor Benchmark`, `Content Planner`	open a live creator workspace around one channel
`Tools`	tabs	`3`	`Single`, `Batch`, `Playlist`	inspect and prepare public YouTube asset downloads
`Deployment`	in-app reference view	`1`	deployment/setup guidance	explain how to run and deploy the app

For the detailed tab-by-tab and module-by-module behavior, see:

Current V6 Architecture In One View

flowchart TD
    A["Bundled GitHub CSVs<br/>data/youtube api data/*.csv"] --> B["streamlit_app.py"]
    U["User actions in the Streamlit UI"] --> B
    B --> C["dashboard/app.py run + st.navigation"]
    C --> D["sidebar navigation"]
    D --> E["7 V6 page views"]

    S["Streamlit secrets / env vars"] --> F["src/utils/api_keys.py"]
    F --> G["YouTube Data API v3"]
    F --> H["Gemini / OpenAI"]

    A --> I["Dataset-backed analysis path"]
    G --> J["Live public-channel and niche-research path"]
    H --> K["AI generation path"]

    I --> L["pandas transforms + visualization helpers"]
    J --> L
    K --> L

    J --> M["Channel Insights service path"]
    M --> M1["public workspace + feature frame"]
    M1 --> M2["heuristic topics or BERTopic beta"]
    M2 --> M3["metrics + outliers + recommendations + snapshots"]

    L --> N["tables, charts, cards, downloads"]
    M3 --> N

The architecture resolves into three repeatable patterns:

Dataset path: GitHub CSVs -> pandas transforms -> benchmark visuals
Live API path: secrets/env -> provider clients -> normalized payloads -> interactive pages
Model path: Channel Insights feature frame -> heuristic or BERTopic topics -> downstream metrics and recommendations

For the deeper page-by-page breakdown, see Architecture.

API And Secrets Flow

flowchart LR
    A["Streamlit secrets / env"] --> B["src/utils/api_keys.py"]
    B --> C["Select / rotate provider key"]
    C --> D["YouTube Data API"]
    C --> E["Gemini"]
    C --> F["OpenAI"]
    D --> G["Channel Insights / Outlier Finder / Ytuber / Tools / Thumbnail URL flow"]
    E --> H["Thumbnails / Ytuber AI Studio / Outlier AI"]
    F --> H
    G --> I["service-layer normalization"]
    H --> I
    I --> J["pandas frames / scored payloads / artifact prep"]
    J --> K["Rendered Streamlit UI"]

This is the live V6 runtime path today:

bundled CSVs power Channel Analysis
live YouTube API calls power Channel Insights, Outlier Finder, Ytuber, Tools, and thumbnail URL export
Gemini/OpenAI power thumbnail generation, AI Studio, and Outlier AI research
the same secret names work in Streamlit secrets or GCP-style injected environment variables

For the deeper API/service explanation, see Architecture.

Channel Insights: Where The Modeling Actually Lives

Channel Insights is where the most advanced modeling work lands in V5/V6. Every refresh starts with the same public-channel workspace, then branches into one of two topic assignment modes:

Heuristic Topics
- default mode
- built from title, tags, and description tokenization
- always available
Model-Backed Topics (Beta)
- optional BERTopic semantic grouping
- activated only when beta mode is selected and the model manifest/artifact path is configured
- falls back to heuristics if anything fails

flowchart TD
    A["Channel Insights UI"] --> B["refresh_channel_insights(...)"]
    B --> C["load_public_channel_workspace(...)"]
    C --> D["ensure_public_channel_frame(...)"]
    D --> E["add_channel_video_features(...)"]
    E --> F["_apply_requested_topic_mode(...)"]
    F --> G["assign_topic_labels(...)"]
    F --> H["apply_optional_topic_model(...)"]
    H -->|artifact missing / invalid / transform failure| G
    G --> I["primary_topic + topic_labels + topic_source='heuristic'"]
    H --> J["model_topic_id + model_topic_label_raw + model_topic_label"]
    J --> K["primary_topic + topic_labels + topic_source='bertopic_global'"]
    I --> L["shared scoring + metrics + outliers + snapshots"]
    K --> L

For the full topic-mode branch, tab flow, and artifact-state details, see Architecture: Channel Insights.

Deployment Summary

Item	Value
Original repo	`matt-foor/purdue-youtube-ip`
V6 development branch	`youtube-ip-v6`
V5 historical branch	`youtube-ip-v5`
Public V6 live app	ip-youtube-creator-insight-2026.streamlit.app
Earlier documented deploy (V5)	`royayushkr/Youtube-IP-V5` on `main` (see Deployment And Versions)
Main file path (Streamlit Cloud)	`streamlit_app.py`
Required secret families	`YOUTUBE`, `GEMINI`, `OPENAI`
Optional secret family	`MODEL_ARTIFACTS_*` for BERTopic beta

flowchart LR
    A["GitHub repo"] --> B["V6 branch / fork"]
    B --> C["Streamlit or GCP environment"]
    C --> D["Secrets / env vars"]
    D --> E["streamlit_app.py"]
    E --> F["dashboard/app.py run + navigation"]
    F --> G["7-page V6 app shell"]

Deployment notes in one screen:

V6 stays public-only for Channel Insights (same as V5)
the app reads st.secrets first, then environment variables
BERTopic beta only activates when MODEL_ARTIFACTS_ENABLED=true and the manifest URL is configured
the GCP-oriented env template is .env.gcp.example, which is meant to be copied into Cloud Run, GCE, or Secret Manager-backed environment configuration rather than committed as a real secret file

For the full deployment matrix and secrets history, see Deployment And Versions.

Project Brief And Evolution Summary

The short version of the project story is:

V1 proved the public-data analytics and recommendation concept
V2 expanded into a broader creator operating system
V3 clarified the product shell and runtime structure
V4 added the deepest intelligence layer with Channel Insights, Assistant, Google OAuth, and BERTopic beta
V5 kept the strongest workflows, removed the heaviest operational complexity, and documented the system clearly for presentation and deployment
V6 is the current Streamlit release: same V5 capabilities, with a hardened router, dependencies, and deployment path (live app)

What V5 removed on purpose (still true in V6):

sidebar Assistant
Google OAuth and owner-only analytics overlays
heavier mixed recommendation behavior on page 3

What V5 kept on purpose (still true in V6):

dataset benchmarking
public tracked-channel insights
thumbnail generation and export
outlier research
the live Ytuber workspace
optional BERTopic beta modeling

What V6 adds on top of V5:

dependable local and Cloud runs (correct Python wheels, PyArrow, Google client stack)
Streamlit-native multipage navigation and a single run() entry from streamlit_app.py so every rerun executes the router

For the full narrative brief and retrospective, see Project Brief.

Where To Read Next

Architecture for the full runtime pipeline, page map, and topic-model integration details
Deployment And Versions for branch targets, secrets evolution, and version/deployment comparisons
Project Brief for the narrative project story, original goals, what changed, and how V5 led to V6

Ethics And Operating Principles

The original V1 principles still apply in V6:

use public data responsibly
respect provider terms of service
avoid exposing personal data
prefer explainable insights over black-box claims
make AI-generated outputs additive to analysis, not a replacement for it

Local Run

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
streamlit run streamlit_app.py

Name		Name	Last commit message	Last commit date
Latest commit History 143 Commits
.devcontainer		.devcontainer
.streamlit		.streamlit
dashboard		dashboard
data		data
docs		docs
outputs		outputs
research_archive		research_archive
scripts		scripts
src		src
tests		tests
.env.example		.env.example
.env.gcp.example		.env.gcp.example
.gitignore		.gitignore
CODE_OF_CONDUCT.md		CODE_OF_CONDUCT.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
SECURITY.md		SECURITY.md
packages.txt		packages.txt
requirements.txt		requirements.txt
setup.py		setup.py
streamlit_app.py		streamlit_app.py

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

YouTube IP V6

At A Glance

Current V6 In One Minute

Why This Project Exists

Version Lineage (V1–V6)

Deployed Version History

What Changed Across Versions

What Survived Into V6

What We Tried And Later Removed Or Simplified

Current V6 Product Surface

Current V6 Workflow Map

Current Interactive Surfaces

Current V6 Architecture In One View

API And Secrets Flow

Channel Insights: Where The Modeling Actually Lives

Deployment Summary

Project Brief And Evolution Summary

Where To Read Next

Ethics And Operating Principles

Local Run

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

YouTube IP V6

At A Glance

Current V6 In One Minute

Why This Project Exists

Version Lineage (V1–V6)

Deployed Version History

What Changed Across Versions

What Survived Into V6

What We Tried And Later Removed Or Simplified

Current V6 Product Surface

Current V6 Workflow Map

Current Interactive Surfaces

Current V6 Architecture In One View

API And Secrets Flow

Channel Insights: Where The Modeling Actually Lives

Deployment Summary

Project Brief And Evolution Summary

Where To Read Next

Ethics And Operating Principles

Local Run

About

Resources

License

Code of conduct

Contributing

Security policy

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages