✨ Smart Feeds

My curated RSS feeds scored by Gemini AI + community signals (HN/Lobsters)

Gemini 3 FlashApache AirflowCloud RunBigQueryFastAPI
Smart Feeds Architecture

πŸ€– Feed Assistant

Natural language search powered by Gemini intent extraction (NL2SQL).

Gemini 3 FlashApache AirflowNL2SQLBigQueryFastAPISemantic SearchVector Searchtext-embedding-004

Gemini 3 Flash Agent

Hello! I'm your Feed Assistant. I can answer questions about the articles in your feed or help you find specific information. How can I help you today?

πŸŽ™ Lecture Caddie

Your AI study companion for technical lectures. Pause the video β€” Caddie explains what was just covered, answers your questions, and surfaces related reading from your feed.

Gemini 2.5 Flash Gemini Live 2.5 Real-time Voice Knowledge Graph Contextual Research WebRTC Daily WebRTC Pipecat ffmpeg BigQuery GCS Redis 24h continuity vector section grounding temporal anchoring
Lecture Caddie Architecture
Real-Time Voice Tutor
Session Lifecycle
Ask Caddie calls /api/lectures/{id}/copilot/start, which provisions a private Daily WebRTC room and mints separate user and bot owner tokens. The Pipecat bot joins via the VM sidecar (/start/lecture) with lecture_data carrying the full sections[] manifest and concept_graph. Gemini Live 2.5 handles bidirectional audio natively β€” no intermediate transcription step.
Grounding & Memory
At startup, each section[] title is embedded and vector-searched against articles_scored in a background precompute task β€” top 3 feed articles per section are cached in memory, injected deterministically when the user pauses on that section. YouTube currentTime anchors every response to the active temporal section. Sessions persist via Redis: on disconnect, Gemini Flash summarizes the conversation into a structured continuity packet {learned, struggling, open_threads} scoped to (article_id, client_id) with a 24h TTL β€” loaded as model context on the next session.

β˜• Morning Briefing

AI anchor delivers your briefing - say "next" for the next article, "read more" for the full article, or ask questions anytime!

Gemini 3 LivePipecatApache AirflowDaily WebRTCWebRTCVoice Activity DetectionReal-time

πŸ“ Today's Script

Loading...
Ready
Tap play to start briefing
Fetching the latest briefing for you...

* Daily briefing is generated between 6–7 AM MST each morning.

βš™οΈ Administration

πŸ€– Model Compatibility

Powered by Gemini

Feature Model Location
Smart Feeds (scoring) gemini-2.5-flash us-central1
AI Assistant (NL2SQL) gemini-2.5-flash us-central1
Lecture Notes (processing) gemini-2.5-flash us-central1
Lecture Caddie (live) gemini-live-2.5-flash-native-audio us-central1
Morning Briefing (script) gemini-2.5-flash us-central1
Morning Briefing (audio) gemini-live-2.5-flash-native-audio us-central1
Code Review (analysis) gemini-2.5-pro us-central1
Code Review (embeddings) text-embedding-005 us-central1

πŸ” Workspace Unlock

Enter your Admin API Key to unlock video processing and release management features.

πŸ”’
Locked: Enter key to proceed

πŸŽ₯ Process New Video Lecture

Generate notes and slides from a YouTube technical talk.

πŸ”„ Data Management

Sync feed definitions from OPML source file.

πŸ” Observability

Live telemetry for cost, latency, and system health

RAG ArchitectureDatadogGCPCloud RunVertex AILLM Ops
Briefing Service Workflow

πŸŽ™οΈ Briefing Observability

Live session metrics from the Pipecat sidecar (TTFA, turn duration, tokens).

β€”
TTFA (P50)
β€”
TTFA (P95)
β€”
Turn (P50)
β€”
Turn (P95)

πŸ“Š Scoring Observability

Airflow DAG metrics from llm_observability.scoring_batch_metrics.

β€”
DAG Runs (7d)
β€”
New Articles
β€”
Stored
β€”
Avg Score
β€”
Low Score Rate
β€”
Community Coverage
See the Smart Feeds Architecture in the ✨ Smart Feeds tab.

πŸ”΄ Code Review Observability

Source: Live telemetry from llm_observability.metrics in BigQuery.

LLM Observability Architecture

Recent Repo Activity

Fetching latest commits...

...
Total Reviews
...
Total Cost
...
Avg Latency