Multi-Brand AI Ops Suite — Creative Studio, Analytics & Copilot (NDA)
End-to-end platform built for creative agencies and marketing teams (client identities protected under NDA): unifies hybrid AI vision, high-volume asset curation, omnichannel BI, and automated campaign production.
- Role
- Product Architect, AI Automation Engineer & Full-Stack Developer
- Material / Stack
- Python · Streamlit · Gemini 2.0 · Ollama Vision · SQLite · MongoDB · Google/Meta APIs
- Process / Engineering
- Hybrid Cloud/Local orchestration, parallel batch processing, OAuth normalization & executive BI
Tools
Deliverables
Web/desktop command center, concurrent visual indexer, content studio & PDF/XLSX exporter
400+ / 50%
Assets / Batch · Time Saved
Lock-free multi-PC parallel partitioning with automated SQLite merge
Cloud + Local
Hybrid AI Engine with Fallback
Gemini 2.0 Flash cloud + local Ollama (moondream/llava) at $0/token
5 Fuentes
Unified Omnichannel BI
Meta/IG, Meta Ads, Google Ads, GA4 & Google Business Profile with zero token exposure
100% NDA
Multi-Brand Isolation
Brand DNA, tone, Kanban boards, and credentials isolated per client
Visual Documentation & Diagrams
01 / Brief & Challenge
Multi-brand agencies and marketing teams lose dozens of hours weekly manually sifting through hundreds of raw photos, stitching disconnected metrics from Meta, Google Ads, and Analytics, and drafting campaigns without brand consistency or AI cost control.
02 / Engineering & Process
Engineered a modular Python/Streamlit suite capable of operating across diverse verticals (music/entertainment, tourism/experiences, and consumer brands) under strict data isolation. Combines a hybrid vision engine (Gemini + local Ollama) with multi-PC concurrent partitioning, automated copy/overlay studio, OAuth-normalized BI connectors, and direct Excel/PDF exports.
03 / Outcome & Impact
Production-deployed operations suite (client identities anonymized under NDA) cutting visual curation cycles in half, eliminating redundant token spend on bulk vision tasks, and grounding campaign recommendations in real performance KPIs (CTR, CPC, CPA, ROAS).
04 / TECHNICAL ARCHITECTURE & SUBSYSTEMS
High-Volume Visual Curation & Brand-Fit Scoring (400+ Assets)
Instead of manually sorting massive photo folders, the indexer generates lightweight thumbnails (Pillow), extracts EXIF metadata (piexif), scores each image against the client's brand DNA, and outputs a numeric fit score with written rationale, tags, and optimal channel format (Post, Story, Reel, Cover) in SQLite and Excel.
Hybrid Cloud + Local AI & Multi-PC Parallel Processing
Quota-resilient architecture pairing cloud Gemini 2.0 Flash with local Ollama vision models (moondream / llava) under an explicit token budget guard. A concurrent partitioning mode splits large media batches across multiple workstations (odd/even isolated SQLite DBs) and merges results cleanly with a single --merge pass.
Content Studio, Logo Detection & Kanban Editorial Grids
Turns scored media directly into ready-to-ship campaigns: auto-detects optimal logo placement, applies consistent brand overlays, generates copy tuned to each account's voice via modular skills, and organizes deliverables across Kanban boards and editorial calendars.
Zero-Token-Exposure Omnichannel BI & Metric-Backed Copilot
Unifies Meta/Instagram, Meta Ads, Google Business Profile (metrics & reviews), Google Analytics 4, and Google Ads through normalized pipelines (Make.com OAuth bridge + structured sheets and direct connectors). The AI advisor cross-references spend, reach, CTR, CPC, CPA, and ROAS to deliver prioritized, metric-backed recommendations.
Operational Ecosystem: Calendar, WhatsApp, MongoDB & PDF
Connects strategy to daily execution: Google Calendar sync, WhatsApp notification gateway, dual local/cloud persistence (JSON, SQLite, Google Sheets & MongoDB), meeting/MCP tooling, and automated executive PDF report generation ready for leadership or client sign-off.
Security, Data Governance & Automated Test Suite
Engineered for real third-party accounts: secrets strictly isolated in .env, per-client data segregation (cliente_id), zero credential exposure in UI or logs, and an automated diagnostic test suite verifying AI providers, syntax integrity, and content studio pipelines prior to release.
05 / SYSTEMS DESIGN & PROGRESSION ECONOMY
| Operational Area | Legacy Bottleneck | Implemented Solution | Commercial Value |
|---|---|---|---|
| Visual Curation | Manually sorting 400+ raw photos without consistent brand scoring. | SQLite indexer + hybrid AI vision with rationale, tags, and multi-PC parallel split. | 50% faster batch turnaround and consistent brand-aligned asset selection. |
| AI Cost Control | Full reliance on paid cloud APIs and frequent rate-limit bottlenecks (429). | Dual Cloud Gemini + Local Ollama orchestration with explicit token budgeting. | Bulk catalog processing at zero local token cost while reserving cloud AI for strategy. |
| Ads & Web BI | Metrics scattered across 5 dashboards with risky manual token sharing. | OAuth-normalized ingestion (Meta, IG, GA4, Google Ads, GBP) + KPI-grounded advisor. | Unified ROI/ROAS visibility and prioritized actions backed by real metrics. |
| Client Delivery | Reports and content calendars stitched manually across disconnected files. | Content Studio + Kanban + one-click Excel (.xlsx) and executive PDF exports. | Standardized executive deliverables ready for immediate client sign-off. |
06 / MODULAR SOURCE CODE MAP
| Module / Path | Architectural Role | Implementation & Detail |
|---|---|---|
| Core & Desktop App (app.py, desktop_app.py, core_engine.py) | Core Orchestration & Operational UI | Streamlit/desktop entrypoint, campaign planning logic, budget calculations, and multi-account workflow orchestration. |
| Hybrid AI & Studio (ai_copilot.py, content_studio.py, token_budget.py) | Hybrid Generative Engine & Content Studio | Cloud/Local routing (Gemini + Ollama), automatic logo placement, overlays, vertical-specific skills & token budget enforcement. |
| Concurrent Vision Indexer (image_indexer.py + SQLite + Pillow) | Parallel Media Catalog Classification | Thumbnail generation, EXIF parsing, brand-fit scoring, odd/even multi-PC partitioning, and lock-free database merge. |
| Omnichannel Connectors (connectors.py, meta_api.py, google_*_api.py, marketing_data.py) | Normalized Marketing Ingestion | Daily metrics, campaign performance, and review ingestion from Meta, Instagram, Google Ads, GA4, and Google Business Profile. |
| Persistence, Sync & Exports (data_manager.py, mongo_manager.py, sheets_manager.py, pdf_exporter.py) | Persistence, Sync & Reporting | Hybrid persistence (SQLite, MongoDB, Sheets), Google Calendar sync, WhatsApp gateway, and executive PDF/Excel exports. |