SDECB

SDECB AI Program

Bilingual Knowledge Hub & AI Assistant for SDECB

Client: SDECB (Societe de developpement economique de la C.-B.) Project Lead: Kris Krug (The Upgrade AI) Status: Implementation Phase Active (Jan-Mar 2026)

Quick Start

Open Portal v2 →

User Guide →

# Run locally with assistant API
python3 scripts/assistant_server.py
# Open http://localhost:8000/portal/v2/

What’s Live Now (January 2026)

Feature	Status	Description
Portal v2	Live	Bilingual knowledge hub with 9 pages
Knowledge Base	Live	2,080+ searchable items (podcasts, interviews, website)
Assistant	Internal Pilot	Source-grounded Q&A with citations
External Chatbot Prototype Plan	In Progress	Spec, governance, rollout approach (Phase 3 build is stop/go gated)
User Docs	Complete	Comprehensive guides for all features
Training	Completed	January 19, 2026 session delivered

Portal v2 Pages

Dashboard - Quick stats and navigation
Knowledge Base Search - Full-text search with filters
Assistant - Q&A with cited sources (internal pilot)
Insights - Pattern analysis across content
Podcast Explorer - Browse 67 episodes by tags
Governance Hub - Policy framework resources
Training Hub - Educational materials
Survey - Data discovery survey for staff
Careers - Job postings (accessible via footer)

Technical Status

Portal v2 (`portal/v2/`)

9 pages: Dashboard, KB Search, Assistant, Insights, Podcasts, Governance, Training, Survey, Careers
Bilingual (EN/FR) with language toggle
Static deployment on GitHub Pages
Modular component architecture (/components/)

Portal Design System (`portal/css/`, `portal/icons/`)

Brand CSS (sdecb-brand.css) - Centralized SDECB brand variables and common styles
Icon Pack (icons/icons.svg) - 40+ custom duotone SVG icons (Red + Navy)
Light theme with SDECB Red (#ED1C2E) accents
Official SDECB logo in header

Backend (`scripts/`)

assistant_server.py - Local dev server + /api/chat endpoint
kb_retrieval.py - Keyword-based search with citations
Retrieval-only (LLM synthesis planned for future)

Knowledge Base (`knowledge_base/`)

2,080 indexed items across 3 sources
Podcasts: 67 episodes with transcripts
Website: 949 pages
Interviews: 24 items (12 staff interviews + summaries)
File-based JSON indexes (no database required)

Ingestion Pipeline

ingest_podcast_transcripts.py - Podcast processing
ingest_website_content.py - Website extraction
ingest_interviews_kb.py - Interview indexing
build_unified_kb_index.py - Master index generation

Project Background

SDECB (Societe de developpement economique de la C.-B.) is a bilingual (French-English) economic development organization serving the francophone community in British Columbia. The organization provides business advisory services, employability programs, and community development initiatives.

Discovery Phase Findings (Aug-Sep 2025)

Through comprehensive staff interviews (12 of 12 completed), we identified:

2,652+ recoverable hours annually (1.3 FTE equivalent)
$92,820 - $132,600 annual value in time savings
20+ automation opportunities across reporting, translation, and client FAQ handling

Key pain points addressed:

Report generation burden (10+ hours/week per manager)
Translation workload (6-8 hours/week organization-wide)
Repetitive client questions (40% of advisor time)

Full Discovery Report →

Project Timeline

Discovery Phase (Aug-Sep 2025) - COMPLETED

12 staff interviews across all organizational levels
Pain point quantification and opportunity mapping
Live demonstration at September staff offsite
7 pilot volunteers recruited

Implementation Phase (Jan-Mar 2026) - ACTIVE

Budget: Ministry-funded (secured December 2025)
January: Training sessions, AI policy co-creation
February: Knowledge base development, external chatbot prototype planning
March: Internal chatbot deployment, external chatbot prototype plan finalized, Phase 3 stop/go gate decision

Documentation

For Users

User Guide - Getting started (15-20 min)
Search Guide - Effective querying (10-15 min)
Graph Guide - Visual exploration (10-15 min)
FAQ - Common questions

For Technical Teams

For Stakeholders

Directory Structure

SDECB/
├── portal/
│   ├── v2/                    # Current production portal
│   │   ├── index.html         # Dashboard
│   │   ├── knowledge_base.html
│   │   ├── assistant.html
│   │   ├── insights.html
│   │   ├── components/        # Shared JS modules
│   │   └── data/              # JSON indexes for static serving
│   └── archive/               # Historical demos (see archive/README.md)
│
├── knowledge_base/
│   ├── unified_index.json     # Master index (2,080 items)
│   ├── podcasts/              # Podcast index
│   ├── interviews/            # Interview index
│   ├── website/               # Website content index
│   └── _sources/              # Raw source content
│
├── scripts/
│   ├── assistant_server.py    # Dev server + API
│   ├── kb_retrieval.py        # Search & citations
│   └── ingest_*.py            # Data ingestion scripts
│
├── docs/
│   ├── user-docs/             # End-user guides
│   ├── technical/             # Developer docs
│   ├── implementation/        # Phase 2 planning
│   ├── discovery/             # Discovery phase report
│   └── governance/            # Policy frameworks
│
├── interviews/                # Staff interview transcripts
├── podcasts/                  # Podcast episodes & metadata
├── README.md                  # This file
├── ROADMAP.md                 # Future improvements
└── CHANGELOG.md               # Project milestones

Historical Demos (Discovery Phase)

These Python demos were built for the September 2025 presentation:

Demo	Purpose	File
Bilingual Voice Assistant	Real-time FR/EN voice interaction	`portal/bilingual_voice_assistant.py`
Report Narrator	Automated reports with voice	`portal/report_narrator.py`
Digital Twin	Interactive Q&A chatbot	`portal/archive/matthieu_digital_twin.html`

See portal/archive/README.md for complete archive documentation.

ROI Summary

Conservative Scenario (Implementation Phase)

Investment: Ministry-funded
Time Savings: 36-38 hours/week in Phase 1
Target ROI: 400%+

Key Metrics to Track

Hours saved per week (target: 36-38)
User adoption rate (target: 80% within 3 months)
API response time (target: <2 seconds)
Bilingual quality assessment

Contact

Project Lead: Kris Krug Organization: The Upgrade AI Client Contact: Mylene Letellier (CEO, SDECB)

Last Updated: January 23, 2026 Recent Milestones:

Training session completed (January 19, 2026)
Data discovery survey deployed (January 23, 2026)

This project demonstrates human-centered AI discovery. By listening first, we identified automation opportunities worth 1.3 FTE equivalents with 400%+ ROI - and earned staff buy-in through transparency, evidence, and respect for organizational culture.