Spaces:
Running
Running
Uploaded new Version of AgileAdvisor v0.5
#2
by Suryasticsai - opened
feat: AgileAdvisor v0.5 — Persistent Gist Sync, ScrumLens RAG, Auto-Tags, Duplicate Detection, Role Hierarchy & Quick-Feed
Summary
This PR ships AgileAdvisor v0.5, a major feature drop that turns the static knowledge base into a living, self-improving RAG system. Every feed now auto-persists back to GitHub Gist, facts carry smart category tags, and a new ScrumLens mode lets users audit raw meeting transcripts against the vector memory.
✨ What's New
| Feature | Description |
|---|---|
| 🔄 Persistent Gist Sync | Every feed—bulk or single-click—now appends to the GitHub Gist API and immediately hot-reloads the ChromaDB collection. The Gist is the single source of truth; local vector DB is just a fast mirror. |
| 🔍 ScrumLens Native RAG | New "ScrumLens 🔍" tab. Paste any meeting transcript (Daily Standup, Retro, Planning) and the advisor cross-checks it against the KB, surfaces anti-patterns, assigns per-role action items, and gives a health score banner (🟢🟡🔴). |
| ⚠️ Duplicate Detection | Live duplicate guard before you save. Uses exact match + difflib.SequenceMatcher (>88% similarity) + substring containment. The Feed tab now has a "Preview Tags & Duplicates" button that renders inline warnings. |
| 🏗️ Role Hierarchy Diagram | New "Role Hierarchy 🏗️" tab. A pure HTML/CSS vertical flow diagram showing Stakeholders → PO → SM → Dev Team → RTE with color-coded borders and assigned output badges (e.g., Product Backlog, Impediment Removal, Potentially Shippable Increment). |
| 🏷️ Feed Data Auto-Tags | Facts are auto-labelled on ingest via keyword heuristics (ceremony, role, artifact, metric, anti-pattern, gamification, general). Tags are embedded in Gist lines as [ceremony] fact... and stored in ChromaDB metadata for richer retrieval context. |
| 📦 Smart Quick-Feed Card | After every answer in the Ask tab, a compact card appears showing the top retrieved fact + its tags. One "➕ Feed This Fact to Dataset" button instantly appends it to Gist without re-typing. |
🏗️ Technical Changes
KnowledgeBase.auto_tag()— zero-token keyword heuristics for categorization.KnowledgeBase.check_duplicate()— exact / similar / contained tri-check.KnowledgeBase._parse_line()/_format_line()— round-trip serialization of[tag] contentformat in Gist.KnowledgeBase._append_lines_to_gist()— unified write path used by both bulk feed and quick feed.AgileAdvisor.analyze_transcript()— transcript audit pipeline with expanded context window (n_results=3).- ChromaDB metadata — now stores
tags,hash, andsourceper document. - Gradio UI — 5 tabs: Ask, ScrumLens, Feed Dataset, Role Hierarchy, About.
🔄 Backwards Compatibility
- Zero breaking changes. Existing Gist lines without a
[tag]prefix are gracefully parsed asgeneral. - The ChromaDB collection version metadata is bumped to
2.0but re-initializes automatically on first run.
🧪 How to Test
- Ask a question → verify the response still renders the Dataset Card + explanation + quote.
- Check Quick-Feed → after the answer, click "➕ Feed This Fact to Dataset" and confirm the Gist grows by one tagged line.
- Open ScrumLens → paste a fake Daily Standup transcript → verify you get a health score + red flags + role actions.
- Feed Dataset → type two lines, one duplicate of an existing fact → click Preview → confirm the duplicate is flagged in red. Then click Feed and confirm Gist sync message.
- Role Hierarchy → open the tab → verify the vertical flow diagram renders with correct role colors and output badges.
- Restart the Space → confirm the KB reloads from Gist with tags intact (check the stats bar document count).
📁 Files Added
AgileAdvisor-v0.5.py
✅ PR Checklist
- Code runs locally without errors
- Gist read/write tested with live token
- ChromaDB persistent storage verified across restarts
- No hardcoded secrets or API keys exposed
- HTML output sanitized via
sanitize_html() - Backwards compatible with untagged legacy Gist data
Ready for review - app.py. 🚀
Suryasticsai changed pull request status to merged