STRAŦUM: The 9-Agent Marketing Application I Built in 75 Days (Solo, While Sick for 10)

Remember that September post where I casually mentioned I was speed-running a 10-agent marketing platform while taking Sunday naps? Four weeks in, 3 agents working, targeting October for alpha?

And that October debugging nightmare where I revealed the name STRAŦUM and mentioned having 8 of 9 agents built?

Well, it’s November now. Time to actually launch this thing.

Table of Contents

The Status:

– ✅ Platform name: STRAŦUM (Intelligence Over Execution)

– ✅ 9 of 9 AI agents built and integrated

– ✅ Multi-tenant architecture for both SMEs and agencies

– ✅ Complete brand guidelines and design system

– ✅ Private-alpha testing phase: NOW

Marketing Execution Without Strategy Is Just Expensive Noise

Most platforms promise faster execution. But speed without direction just burns budget faster.

STRAŦUM is different: 11 strategic frameworks. 9 AI agents. Intelligence that grows with every conversation.

Here’s what that actually means:

11 Strategic Frameworks Applied to YOUR Business:

SWOT Analysis, Porter’s Five Forces, Blue Ocean Strategy, BCG Matrix, VRIO, McKinsey 7S, OKRs, Three Horizons, ICE Prioritization, Business Model Canvas, Jobs to Be Done.

Progressive Learning – Tell One Agent, Inform All Nine:

Unlike static tools, STRAŦUM’s AI agents continuously learn about your business with every conversation. The more you interact, the smarter and more targeted the insights become.

How It Works:

Day 1 – Automatic Insight Capture:

Share your business context once. Agents provide strategic frameworks and begin automatically capturing key insights about your market, competitors, and customers. No clicking “Save” – intelligence accumulates naturally.

Week 1 – Cross-Agent Intelligence Sharing:

Every conversation adds to your Learning History. Talk about European expansion with the Strategy Agent? The Content Agent already knows about it next week. Agents reference past insights to provide increasingly personalized recommendations. No re-explaining context.

Ongoing – Predictive Intelligence:

With a rich knowledge base, agents anticipate your needs and pre-fill contexts. High-confidence insights (≥90%) are automatically approved. Lower-confidence learnings wait for your review. You stay in control – view your complete Learning History and delete any insight you don’t like.

5-Minute Quick Wins:

Actionable insights in your first session, not your fifth week.

The Evolution: From DIALØGUE to STRAŦUM

Building DIALØGUE taught me how to ship AI products. Building STRAŦUM taught me how to build platforms.

DIALØGUE (8 months, Jan-Aug 2025):

One user type: Individual podcast creators

One workflow: Research → Script → Audio generation

14 microservices: Lambda → Cloud Run migration

Simple auth: JWT for one user type

One revenue stream: Credit packs ($4.99-$19.99)

Business model: B2C, single-tenant

STRAŦUM (75 days, Aug-Nov 2025):

Two user types: SMEs + Agencies (managing 5-15 clients each)

9 specialized agents: Each with multiple tools, sharing intelligence

45+ database tables: Complete multi-tenant data isolation

Complex auth: Organization → Client → Campaign hierarchy

Business model: B2B + B2C, multi-tenant SaaS

Why Multi-Tenant Architecture Is 10X Harder

DIALØGUE generated podcasts for one user at a time. STRAŦUM manages marketing intelligence for agencies juggling multiple competing clients.

The Challenge: An agency managing multiple clients needs:

Complete data separation (Client 1 can’t see Client 2 data)

Hierarchical context (organization → client → campaign)

Cross-agent intelligence sharing (within campaign boundaries only)

Schema routing (`public` schema for SME, `agency` schema for agencies)

Row Level Security on 45+ database tables

Example: When a strategist uses the Business Strategy Agent for Client 1, that analysis is isolated. Switch to Persona Agent? It pulls Client 1’s strategy – but CAN’T see Client 2. Brand guidelines cascade down to Content Agent, ensuring every piece uses Client 1’s voice, not Client 2’s.

This required database functions, materialized views, trigger-based real-time updates, and honestly way more SQL than I ever wanted to write.

200+ commits over 72 days just for multi-tenant architecture. That’s not a feature – it’s an architectural philosophy that touched every layer of the stack.

The Numbers (Because I Can’t Help Myself)

75 days. August 20th to November 3rd. Here’s what it took:

Development Velocity:

Git commits: 1,000+ (actual count: 1,075) – that’s 14.5 commits/day average

Lines of code: ~200,000 (Python: 62k, TypeScript: 98k, SQL: 41k)

Database migrations: 214 sequential migrations

Agents built: 9 of 9 (all core agents shipped)

Technical Complexity:

Database tables: 45+ tables with complete RLS policies

RLS policies: 83 policies across 26 tables for multi-tenant security

Foreign key indexes: 98 indexes added (Postgres doesn’t auto-create them!)

Color token migration: 700+ instances across 200+ files in one day

Major architectural pivots: 3 (ADK→Direct API, Nuclear Migration, Database-First)

Performance Improvements:

Latency reduction: 72% faster AI responses (hybrid function calling)

RLS optimization: 10-100x query speedup with policy caching

Bundle size: 92% reduction through code splitting

Reality Check:

Navigation bugs from multi-tenancy: 23 (fixed in 2 days)

Days lost to sickness: 10 (still shipped on time)

Days on vacation: 8 (beach ≠ debugging)

Coffee consumed: Still don’t ask

Times I almost gave up: 0 😛

Times Claude Code / Gemini 2.5 Pro saved me: Honestly lost count

What I Learned (Again)

1. Multi-Tenancy Is Hard

Data isolation isn’t just adding `org_id` to every table. It’s thinking through:

– Which schema does this data live in? (`public` for SME, `agency` for agencies)

– What happens when you delete a campaign? (Soft delete with `archived_at`, not hard delete)

– How do permissions cascade? (Organization admin vs client manager vs campaign contributor)

Example: On November 1st, I fixed 23 navigation bugs in one day. The issue? Agency users navigating between clients broke URL context. SME routes look like `/persona/session/123`, but agency routes need `/clients/[client-slug]/agents/persona/session/123`. Every agent page needed refactoring to preserve client context across navigation.

That’s 200+ commits over 72 days just for multi-tenant architecture. Not a feature – an architectural philosophy that touched every layer of the stack.

2. 10 Days of Sickness Breaks Momentum

That October launch I mentioned? Yeah, I got sick. Couldn’t look at screens. Couldn’t code. Just had to… wait.

Solo development means there’s no team to pick up slack. But it also means no pressure to ship before you’re ready. I chose to get it right over getting it fast.

3. AI-Assisted Development Is Real (But Not Magic)

That September speed-run wasn’t exaggeration. Claude Code and Gemini CLI let me ship architectural refactors in hours that would’ve taken days.

Example from September 14th:

“`

08:04 AM – Migrated frontend to standardized API client

11:34 AM – Centralized route configuration (no hardcoded URLs)

1:00 PM – Standardized all 10 agent pages

4:38 PM – All agents integrated with context system

5:03 PM – Testing & Polish (92% bundle size reduction)

“`

Six major features. One Sunday. While going to church, grocery shopping, having lunch, taking a nap, and playing iPad games.

But here’s the thing: the AI didn’t write the architecture. It didn’t decide on multi-tenancy patterns. It didn’t debug that [HTTP/HTTPS nightmare](https://www.chandlernguyen.com/blog/2025/10/21/http-https-debugging-nightmare/). It *amplified* my decisions. Thought partner, not ghostwriter.

The 3 Pivots That Shaped STRAŦUM

Building this platform required three major architectural decisions that fundamentally changed the trajectory:

Pivot 1: Direct Gemini API (Day 2 – August 21)

On day 2, I abandoned Google’s ADK (their Gemini SDK wrapper) for direct API access. ADK had session management limitations that conflicted with multi-tenant architecture. The migration took 24 hours. Early pivots are cheap. Late pivots are expensive.

Pivot 2: Multi-Tenant from Day 1

I could’ve built for SMEs only. Instead, I chose to support agencies managing multiple clients. This decision added 3 months of complexity: schema routing, data isolation, client context propagation. But it also opened enterprise sales potential – agencies managing 5-15 clients pay proportionally more than individual businesses. 10X revenue per customer justifies 3X development time.

Pivot 3: Nuclear Migration (October 11-22)

By October, I had 9 separate intelligence tables (one per agent type). Each new agent required new migrations, new API endpoints, new frontend queries. I consolidated all 9 into one unified table with flexible schema-less content. The migration took 11 days. Now adding new agents takes hours, not days.

These weren’t technical failures – they were strategic decisions. The AI helped me execute faster, but the architecture decisions were mine.

What’s Working

9 AI marketing agents applying 11 strategic frameworks

Multi-tenant data isolation – agencies can manage multiple clients safely

Progressive learning system – cross-agent intelligence sharing within campaigns

Real-time SSE streaming for all agent conversations

Interactive persona interviews that capture nuanced customer insights

Marketing strategy bridging business strategy to tactical execution

This is a working platform, not vaporware. Private alpha means real users are already testing it.

Private Alpha: Request Early Access

STRAŦUM is live and accepting early testers through invitation-only access. I’m looking for:

Small businesses or startups (1-10 people) who need strategic marketing intelligence

Marketing agencies managing multiple clients who want efficient strategy tools

– Early adopters who want to shape the product through feedback

Request access here. I personally review each request and grant access within 24-48 hours.

What you get:

– 9 AI marketing agents with 11 strategic frameworks

– Multi-campaign management (agencies: manage multiple clients)

– Progressive learning system that gets smarter with every conversation

– Direct access to me for feedback and feature requests

What to expect:

– Private alpha = actively evolving based on user feedback

– I’m responsive to bugs and feature requests

– Solo founder = authentic, hands-on support

The Bigger Vision

After 20 years in advertising, I’ve seen the same pattern: great marketing strategy is expensive and inaccessible. Agencies charge five figures per month. Good strategists cost six figures per year. Solo founders and small teams get left behind.

But what if strategic marketing intelligence could be augmented by AI? Not replaced – augmented. Where AI handles the frameworks, the research, the structured thinking, and humans bring the creativity, the intuition, the je ne sais quoi that makes marketing actually work.

That’s STRAŦUM. Intelligence over execution. Strategy over tactics. Thought partner over ghostwriter.

Is it perfect? Hell no. Is it useful? I genuinely think so.

Final Thoughts (Or: Why I Keep Building in Public)

Building STRAŦUM has been harder than DIALØGUE. More complex. More expensive. More nights where I questioned whether anyone would actually want this.

But here’s what keeps me going: the same reason I built DIALØGUE. Because I wanted it.

And because documenting this journey – the victories, the debugging nightmares, the 23 navigation bugs, the 3 architectural pivots, the 214 database migrations – helps other solo builders see what’s possible.

75 days ago, building a 9-agent marketing platform with multi-tenant architecture seemed impossible for one person. Today, I’m inviting people to test it.

What changed? Better AI tools: Claude Code in particular.

Want to try STRAŦUM? Request an invitation.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.