Turns your processes into simple, reliable software, from Paris.

3018, Emma custom CRM

Client

3018

Scope

Emma, all-in-one custom CRM

Integrations

Chat, phone, email, WhatsApp, Messenger

Role

Freelance consultant

The starting point

3018 is the French national hotline for protecting minors online, run by the e-Enfance association. No off-the-shelf tool could keep up. Day to day, listeners juggled several channels and scattered information, and the tool reached its limits when cases turned urgent.

What changed

Today, listeners work on Emma, a single workstation designed for them. Every channel (chat, phone, email, WhatsApp, Messenger) lands in one queue. Distress signals are detected automatically and suicide-risk cases rise to the top. Each case is documented in a structured, exportable file, and supervisors follow activity in real time. Everything is hosted in France, on sensitive data, with a full audit trail.

What it proves

Designing and shipping the custom Emma CRM: its multichannel queue, integrated telephony and sovereign AI, on highly sensitive data.

IA Brew. AI Newsletter

Type

Automated newsletter

Sources

20+ feeds tracked

Frequency

Weekly

Human input

None, from collection to send

The starting point

Keeping up with AI news took hours of reading every week: dozens of sources, plenty of duplicates, little usable signal.

What changed

The newsletter now writes itself. More than 20 sources are collected, duplicates merged, the best stories selected, summarised and laid out, and the email ships every week. Nobody touches anything, and every step is logged so any anomaly can be traced.

What it proves

The same pattern applies to any team monitoring workflow: collect, sort, summarise, distribute. The difference between a gadget and a reliable tool is tracking every step.

Multi-source Job Scorer + ATS PDF Generator

Sources

WTTJ, JobTeaser, LinkedIn

Jobs scored

240+

Auto-generated

ATS CV + cover letter in PDF

Scoring

Weighted to the profile

The starting point

Serious job hunting took hours: sorting offers from three platforms, then tailoring a CV and cover letter for each role. And most CVs are first read by tracking software (ATS) before any human sees them: an overly creative layout sinks an application without anyone noticing.

What changed

The tool pulls offers from the three platforms, scores them on weighted criteria (240 offers analysed so far) and presents everything in a filterable table. For each promising offer, it automatically generates a CV and cover letter tailored to the role, exported as PDFs built to be parsed correctly by tracking software. The sorting that took hours now runs itself; only the decision remains.

What it proves

Data processing is only worth its rigor: if the data is badly structured, the score means nothing. The same standard applies to your business data.

Juice Jacking Guard. USB Threat Monitor (Windows)

Platform

Windows 10 / 11

Threats covered

BadUSB, Rubber Ducky, O.MG Cable, juice jacking, USB payloads

Size

~4,900 lines, 20 modules, 11 subsystems

Distribution

Single-file .exe installer (23.3 MB)

The starting point

A USB device is never harmless. Windows trusts whatever it declares: a fake stick posing as a keyboard can type 200 commands in seconds, a rigged cable can steal data. No antivirus catches it in time.

What changed

With Juice Jacking Guard installed, any unknown device is blocked before it can even speak to Windows, then classified across four risk levels. A new keyboard has to prove a human is behind it: a specific key to press that an automated payload can neither read nor guess. Storage devices are scanned read-only on insertion, with an optional reputation check. Everything stays local, with no telemetry.

Download (Windows 10 / 11)

SHA-256 (.zip): a58fb6cc307b64d7354a84f9a1447a29dd473446df6e418f8dbb5e8711d851c9

SHA-256 (.exe): 22d6c49f73cbb2564d3be8112438ce1553ac6174706432eb9d24157aa2d4b2bf

What it proves

Taking a niche, half-understood threat and shipping a defence that works, down to the installer and a clean uninstall.

Bloomberg-Style Dashboard

Scope

6 assets (crypto, stocks, ETF)

Briefs

3 per day, automatic

Alerts

Telegram, 3 tiers

Schedule

08:50 / 13:00 / 17:30 Paris

The starting point

Following six assets seriously meant a wall of open tabs, indicators to recompute and decisions made in the heat of the moment.

What changed

A single dashboard, one I use every morning, follows BTC, ETH, SOL, NVDA, TTE and CW8. It pulls prices, computes the technicals and pushes three briefs a day to Telegram, at the hours that matter: before the open, at midday, after the close. On top of that: a scenario simulator, a recurring-investment tracker, a market events calendar and a trade journal.

What it proves

Data collection, computation, automated briefs, scheduling: the same foundation as any business dashboard. Only the subject differs.

IchimokuSignal Pro

Type

Indicator + backtester

Version

v3.3 (Bounce + 15% SL)

Net return (AAOI, 5y)

+187% vs +1088% B&H

Win rate / PF / MaxDD

54% / 1.62 / 24%

The starting point

Most Ichimoku indicators fire too often on long-term charts. I wanted a single GO or WAIT signal for stock picking, validated by a backtest that includes commissions and execution costs.

What changed

The indicator combines ten criteria into one composite score, with a 15% protective stop and safeguards against false signals. A backtester replays the logic candle by candle, round-trip fees and slippage included.

Results (AAOI, 5y)

• Strategy: +187%, 54% win rate, max drawdown -24%.
• Buy-and-hold baseline: +1088%, max drawdown -62%.
• So the strategy loses on absolute return, but cuts the downside risk by more than half.

What it proves

A single-stock backtest is always partly overfit; choosing a strategy is choosing a risk profile.

Real Estate Investment Model

Scope

SCI/IS vs LMNP, mortgage

Data

INSEE, notary stats, rental yields

Cities ranked

34 French metros

Purpose

Size my own portfolio

The starting point

Choosing between an SCI at corporate tax and LMNP is too often done on gut feeling, even though the tax wrapper weighs more than the purchase price on the final return.

What changed

A model now compares both regimes for every deal: year-by-year cash flow, 10-year IRR, tax impact, net equity at exit. It includes mortgage scenarios, an index ranking 34 French metros by rental attractiveness, and macro scenarios (rates, inflation). I use it on deals I underwrite.

What it proves

Turning a complex, numbers-heavy decision into a tool that settles it. The same reasoning applies to any structured investment decision.

Crypto Trading Bot (MEXC)

Strategy

Mean reversion, BTC + ETH

Paper P&L

+3.2% over 90 days

Live P&L

-2.1% over 30 days

Win rate

46% (214 trades)

The starting point

Short-term crypto pairs often overshoot their equilibrium. The hypothesis: when two indicators flag an excess at the same time, a snap-back is more likely than a continuation.

What changed

A bot tests that hypothesis in real conditions: a double-confirmation entry filter, capped position sizes, automatic exits, a live results curve and a kill switch. First in simulation, then with a small-scale live pilot.

Results

• Simulation (90 days): +3.2%.
• Live pilot (30 days, 10% of simulated size): -2.1%, 46% win rate.
• The gap comes from fees and execution costs the simulation did not model; fees alone explain roughly 4 points.

What it proves

I also publish what loses. A simulation that ignores friction costs overstates the result; the corrected version is still running, in simulation, until proven otherwise.