{{ b.cat }}
{{ project.name }}
{{ project.role }}
{{ project.desc }}
Pain point solved
{{ project.pain }}
{{ currentSection }}
{{ tokens }} tokens
Loic Gilles
Djayep
Electrical Engineer & Harvard MBA Candidate building with AI
Always building something
Portrait
Cambridge, MA
Studied & worked at
02
Selected work
Below is a selection of projects I have built
Web apps
01
AI Radar
technology watch
02
SIMON
purchase decisions
03
Frugal
personal finance
04
Colori
colour combinations
Agentic systems
06
Marvin
commercial diligence
07
AI Brain
knowledge repository
Deterministic and AI automations
08
Investment workflows
portfolio notes
09
Budget tracking
spending vs budget
10
Tech Intelligence
AI news & new tools
preview
Case study
case_study
Client work
Custom agentic system
Custom AI Agent
for Client Activation
A custom agentic system supporting client activation from client context through proposal development, CXO interviews, facilitation material and synthesis.
Workflow map
reviewed
01
Client
context
context
Organises approved client context.
02
Proposal
Supports proposal development.
03
CXO
interviews
interviews
Structures interview preparation and findings.
04
Facilitation
material
material
Prepares facilitation materials.
05
Synthesis
Produces reviewable workshop artifacts.
Core elements
Claude Agent SDK
Built around scoped subagents and agent workflows.
Node.js agent layer
Runtime for the agent layer and operational scripts.
Scoped connectors
For approved tools and sources.
Workflow orchestration
Connects client context through synthesis.
Reviewable artifacts
Structured materials for workshop work and review.
Why it matters[+]
Workshop work can involve approved context, research and materials that need to remain structured and reviewable across the engagement.
What I developed[+]
A custom system built around Claude Agent SDK, scoped subagents, approved connectors and workflow orchestration.
What it can produce[+]
It produces structured workshop materials and reviewable artifacts for the workflow.
How it works[+]
The Node.js agent layer runs the workflow and operational scripts across the five stages.
What it shows[+]
This project demonstrates how I design AI systems around a real operating workflow: specialist agents, shared context, reusable standards and quality controls.
03
My team of AI Agents
Each agent holds its own brief and its own context
Aldrich
learning
Dora
research
Godin
positioning
Rawlings
knowledge
Gigi
chief of staff · dispatch
Stark
product & code
Crayon
design & UX
Cresus
finance & risk
Saimon
purchasing
Camus
writing
agent
{{ agName }}
{{ agRole }}
{{ agDesc }}
“I use my agent team to keep track of work, stay organised across the different parts of my life, manage investments and maintain the knowledge I rely on.”
04
Background
Professional experience
Strategy Consulting
Africa · 2021—2025
McKinsey & Company
[+]
Engagement Manager. Led strategy, digital and operating-model work for governments and companies across Africa, including internet penetration, retail strategy and commercial due diligence.
Private Equity
Paris · 2020
Amethis
[+]
Private Equity Analyst Intern. Assessed African investment opportunities through market research, financial analysis and due diligence.
Investment Banking
Paris · 2018—2019
BNP Paribas CIB
[+]
M&A Analyst Intern. Built valuation analyses and pitch materials for industrial-sector M&A transactions.
AI Building & Deployment
2026
Keystone
[+]
AI Builder / Deployer · MBA Internship. Builds agentic workflows and reviewable client-facing artifacts with human approval gates.
Education
Harvard Business School
2025—2027
MBA Candidate
ESSEC Business School
2017—2020
Master of Science in Management
ESTP
2014—2017
Electrical & Mechanical Engineering
“Before becoming an AI builder, I developed my business acumen through roles in strategy consulting, investment banking and private equity.”
05
AI credentials
Here is a non-exhaustive selection of the training I have completed
agentic
Agentic development
“Agents Course”
Hugging Face
Agent architecture, tool use and the patterns behind multi-agent systems.
View certificate↗coding
Coding with agents
“Claude Code in Action”
Anthropic
Working with a coding agent end to end: context, planning, review and iteration.
View certificate↗prompting
Prompt engineering
“Prompt Engineering for ChatGPT”
Vanderbilt University · Coursera
Prompt patterns, structured outputs and reliable instruction design.
View certificate↗Community
Harvard Claude Code Competition
Jury Member of the first Claude Code Hackathon at Harvard University.
06
My AI stack
Here is a non-exhaustive selection of tools that I use
Lovabledev
v0dev
Hermes Agentagents
OpenClawagents
Grok Botagents
Obsidianknowledge
Lovabledev
v0dev
Hermes Agentagents
OpenClawagents
Grok Botagents
Obsidianknowledge
“My first rule when building with AI is not to be loyal to any provider. Each has its own advantages and specificities, so the work is to build a system that lets me move between them.”
07
Essays
Read on SubstackAI Tool
Token economics
Reflection