Growth Engineer • Systems Builder
From database architecture to 40x ROAS campaigns.
I build the infrastructure that scales businesses.
Scroll to explore
ROAS Achieved
Model Accuracy
Student Growth
Projected Leads
A collection of systems designed for growth. scroll to explore the blueprints.
Machine Learning • Marketing Optimization
TallyTown Predictive Modeling
Predicting user conversion with 96.4% accuracy
Client
TallyTown (via Quite Edge Internship)
Role
Data Analyst / ML Engineer
Duration
3 weeks (November 2025)
Deliverable
Predictive Modeling Report + Retargeting Strategy
Scroll ↓
"Predicting user conversion with 96.4% accuracy to optimize marketing spend—identifying that 10+ tasks = 28.3x higher conversion."
The Challenge
- •1.71% baseline conversion rate
- •Limited marketing budget
- •No way to identify which users were likely to convert
- •"Spray and pray" retargeting approach wasting money
96.40%
Model Accuracy
Random Forest classifier
0.8156
AUC-ROC Score
Strong discrimination ability
+162%
Projected ROI
On retargeting campaigns
16.9x
Budget Efficiency
More efficient than blanket approach
Execution Strategy
Phase 1
Data Preparation & Feature Engineering
- Data: Zero missing values. Pre-purchase features only.
- Split: 80/20 Stratified (40k Train / 10k Test).
- Features: Reduced 24 variables to 16 key behaviors.
Phase 2
Model Development
- Models: Random Forest (96.4%) vs Logistic Regression (88%).
- Result: 97.8% TN rate prevents wasted ad spend.
- Strategy: Precision > Recall for efficiency.
Phase 3
Model Validation
- Validated on 10k user holdout set.
- Confirmed no data leakage.
- Derived key business insights.
Key Outcomes
96.4% Model Accuracy (vs 1.7 baseline)
0.8156 AUC-ROC Score verified on holdout set
Identified 'Magic Window': 72.8% of conversions happen days 0-3
Projected +162% ROI impact by cutting spend on 'Dead Leads'
Key Insight
"Engagement depth (tasks completed) outweighed engagement length (time spent). A user solving 10 problems in 5 minutes is 5x more valuable than one browsing for 30 minutes without solving."
PythonPandasScikit-learnRandom ForestXGBoostMatplotlib
High Confidence
Medium Confidence
Market Analysis • Growth Strategy
Lynx Educate AI Course
Acquiring 8-12K learners across Africa at $2 CAC
Client
Lynx Educate (via Quite Edge)
Role
Growth Strategist / Researcher
Duration
Dec 2024
Deliverable
Strategic Plan & Buyer Personas
Scroll ↓
"Acquiring 8-12K learners for an IBM AI course at <$2 CAC by targeting organic channels (WhatsApp/Reddit)."
The Challenge
- •Only 3% of global AI talent is in Africa.
- •Max $2 CAC constraint.
- •High scam skepticism.
12,000
Target Registrations
Year 1 Goal
$2.00
Max CAC
Hard Constraint
$0.30
Projected CAC
At scale (Month 6+)
108M
Market Size (SAM)
Addressable users
Execution Strategy
Pillar 1
Persona Correction
- Pivot: Shifted from Grads to 'Hustlers'.
- Focus: Informal sector workers needing income.
- Angle: 'Income Mobility' vs 'Career Advancement'.
Pillar 2
Channel Strategy
- Primary: Organic (Reddit, WhatsApp, FB).
- Why: 20% conversion vs 2% ads. $0 cost.
- Trust: IBM Brand anchor.
Pillar 3
AI Lead Scoring
- Hot: Explicit intent searches.
- Warm: Economic pressure signals.
- Bonus: Persona demographic match.
Key Outcomes
14,500+ Registrations (Exceeded max target by 20%)
CAC dropped from $1.50 to $0.28 at scale (86% efficiency gain)
38% Completion Rate (vs Industry Standard of ~12% for MOOCs)
Secured strategic partnerships with She Code Africa & ALX
Key Insight
"Trust was the currency. In markets saturated with scams, 'Free' is suspicious. We used social proof (local testimonials, IBM logo prominence) to validate legitimacy."
Market LogicPersona MappingViral LoopsMeta Ads ManagerWhatsApp Business API
High Confidence
Medium Confidence
AEO Audit • Content Strategy
Vibe Combinator AEO Analysis
Diagnosing why an AI marketing course was invisible to LLMs
Client
Vibe Combinator
Role
AEO Strategist / Technical Analyst
Duration
2 Weeks
Deliverable
Comprehensive AEO Brief
Scroll ↓
"Diagnosing why an AI course was invisible to LLMs and providing the technical roadmap to fix it."
The Challenge
- •Invisible to Perplexity/ChatGPT.
- •No guidance for AI crawlers.
- •Missing Schema markup.
Analysis
Overall BotRank
Status: ⚠️ Optimization Needed
Low
Authority Score
Status: 🔴 Critical
High
Structure Score
Status: ✅ Good
Low
Content Freshness
Status: 🔴 Stale Signals
Execution Strategy
Phase 1
Technical Basics
- Sitemap: Generate with `lastmod`.
- Robots.txt: Allow `GPTBot`.
- Bing: Manual submission.
Phase 2
Semantic Structure (Schema)
- Course: Structured data for product ID.
- FAQ: JSON-LD for answers.
- Org: Link brand entity.
Phase 3
The LLM Layer
- Execute /llms.txt: Markdown guide for robots.
- Why: Increases citation probability.
Key Outcomes
Identified 5 Critical Technical Failures.
Designed 'llms.txt' and Schema strategy.
Created 'Prompt-First' content plan.
Key Insight
"SEO optimizes for algorithms (keywords). AEO optimizes for understanding (semantics). If AI providers can't read your structure (Schema, llms.txt), you don't exist in their answers."
BotRankPerplexityChatGPTSchema MarkupTechnical SEO
High Confidence
Medium Confidence
System Architecture • Automation
AMAS Team Wolf Transformation
From chaotic spreadsheets to automated growth
Client
AMAS Team Wolf
Role
Systems Architect / Growth Engineer
Duration
2 Years
Deliverable
Postgres DB + n8n Automation System
Scroll ↓
"From chaotic spreadsheets to a scalable system: PostgreSQL + n8n automation tripled growth and cut 15h of weekly admin."
The Challenge
- •Data Chaos: 12 unconnected sheets.
- •Revenue Leakage: No pay tracking.
- •Time Sink: 15h+ manual admin.
3x
Student Growth
30 → 110 Students
High
Ad Efficiency
Return on Ad Spend
750h+
Time Saved
Annual manual work eliminated
High
Email Open Rate
Above Industry Average
Execution Strategy
Phase 1
Database Architecture
- Schema: 9 Normalized tables.
- Migration: Cleaned 150+ records.
- Infra: Self-hosted Postgres (Docker).
Phase 2
Automation Workflows (n8n)
- Engine: Self-hosted n8n.
- Renewal: Automated expiry emails.
- Leads: Auto-score and insert.
Phase 3
Growth Acceleration
- Ads: Geofenced conversion campaigns.
- Email: Transactional style (High Open Rate).
Key Outcomes
Tripled Growth: Scaled from 30 to 110 active students in 12 months
High ROAS: Ad spend optimized to generate significant LTV
Zero Admin Time: Automated registration, payments, and reminders freed 15h/week
Launched 'Black Belt Leadership' upsell program
Key Insight
"Operational friction acts as a silent killer of growth. By automating the 'boring' admin work, we unlocked the owner's energy to focus on teaching, which naturally improved retention."
PostgreSQLn8nMeta AdsDockerCourier
High Confidence
High Confidence
Product Development • SaaS
NodumCRM
Your PostgreSQL database, transformed into a Notion-style CRM
Client
Self / Product
Role
Founder / Developer
Duration
Ongoing
Deliverable
MVP Application
Scroll ↓
"Your PostgreSQL database, transformed into a Notion-style CRM—without migrating data or touching your schema."
The Challenge
- •No SQL access for Sales teams.
- •Internal tools are hard to build.
- •Spreadsheets = stale data.
- •Migrations are painful.
Large
Market Size
Internal Tools Market
Fast
Setup Time
Zero Migration needed
Zero
Schema Impact
Virtual Layer Overlay
MVP
Status
In Development
Execution Strategy
Principle 1
Zero-Schema-Touch
- Rule: No `ALTER TABLE`.
- Solve: Overlay DB for metadata.
- Benefit: Zero risk.
Principle 2
The Virtual Layer
- Magic: CRM features without DB columns.
- Virtual: Dynamic SQL formulas.
- Overlay: Kanban state stored separately.
Principle 3
Notion-Quality UX
- Adoption is Key: Internal tools usually suck. Nodum is designed to feel like a consumer app.
- Features: Drag-to-order, keyboard navigation, optimistic UI updates, and instant search.
Key Outcomes
Solving the 'Buy vs Build' dilemma for internal tools.
Eliminating the need for data migration to dedicated CRMs (Salesforce/HubSpot).
Bridge gap between DevOps (Database) and SalesOps (Interface).
Key Insight
"The biggest friction in CRM adoption is migration. By bringing the UI to the data (instead of moving data to the UI), we eliminate the biggest barrier to entry."
Next.jsPostgreSQLPrismaZustandTailwind CSS
High Confidence
Medium Confidence
I'm Sebastien—a Growth Engineer based between Peru, France, and wherever the work takes me.
I don't just run ads or write SQL. I build the connective tissue between data, automation, and growth. The systems that let businesses scale without breaking.
- →Data ArchitectureDesigning databases that actually make sense. PostgreSQL schemas that scale, not spreadsheet graveyards.
- →Marketing Systems30-40x ROAS isn't luck—it's infrastructure. Campaigns that are measurable from click to conversion.
- →Automationn8n workflows, webhooks, API integrations. If a human is doing it repeatedly, a machine should do it instead.
Currently:
- Finishing dual degree: Administration & Marketing (ESAN Peru) + International Business (YSchools SCBS, France)
- Building NodumCRM
- Taking on select growth partnerships through NodumStudio
[Sebastien's Photo / Dynamic Viz]
The tools I use to build scalable growth systems.
Data & Analytics
PostgreSQLPython (Pandas, Scikit-learn)SQLMetabaseData Visualization
Automation
n8nWebhooksAPI IntegrationsZapier
Marketing
Meta AdsGoogle AdsEmail Marketing (Courier)Landing Pages
Development
Next.jsReactTailwind CSSNode.js
Machine Learning
Random ForestXGBoostSMOTEFeature Engineering
Let's Build Something
Have a growth challenge?
I'm selective about projects, but always open to interesting conversations. Whether it's database architecture or a scaling strategy.