TLP:CLEARPROFILE NOTE · NO. 2026-11 · F. MBOHOU
Subject
Fils Aboubakar Sidik MBOHOU
I analyse network flows and digital artefacts and document verifiable findings. My Data Engineering experience adds a demand for reliable, traceable data.
Available now · Permanent or fixed-term roles · Angers, open to relocation across France
Open to opportunities in cybersecurity consulting, GRC and CSIRT, as well as PhD projects in academia or industry.
176,758network flows analysedThreatLens case file
91.6%of attacks detected16.7% of benign flows flagged incorrectly · 0.48 threshold
1.29%FGSM attack successdown from 2.07% before hardening · D04
16.43/20MSc averageRanked 1st in semesters 2 to 4
Each project handled as a case file: the question, the evidence examined, the finding. Academic projects carried out in authorised labs, alone or in teams, plus one production assignment.
D01 · Forensics · team of 4 · 2025
Reconstruct a scenario from three seized systems: disk, memory and network, examined using a four-step digital-forensics workflow.
Autopsy · FTK Imager · Volatility · Wireshark · HxD
Finding: Quasar RAT traces identified in memory.
CLOSED Read the case file →D02 · Detection · team of 3 · 2024
Spot attacks in 176,758 CSE-CIC-IDS2018 flows: models compared, threshold set at 0.48, errors analysed.
Python · pandas · scikit-learn · Flask
Finding: 91.6% of attacks detected; 16.7% of benign test flows incorrectly flagged.
CLOSED Read the case file →D03 · Network · pair work · April 2025
Authorised lab: understand what a compromised server sitting between two network zones can reach, from Kali to Debian.
Ligolo-ng · Nmap · Linux
Finding: internal host made reachable through a pivot.
CLOSED Read the case file →D04 · AI security · team of 4 · 2025
Test a PyTorch CNN against the FGSM attack (ε = 8/255), then harden it with adversarial training. Public code.
PyTorch · FGSM · GitHub
Finding: attack success rate cut from 2.07% to 1.29%.
CLOSED Read the case file →D05 · Detection · individual · 2025
Filter 5,157 unique messages without blocking real ones: a sequential BiLSTM model compared with a TF-IDF baseline.
Python · Keras (BiLSTM) · scikit-learn
Finding: 98.32% accuracy, spam F1 0.930; 4 legitimate messages blocked out of 678.
CLOSED Read the case file →D06 · Supporting evidence · final-year internship
Optimise a production PostgreSQL warehouse: ETL cycle cut from 30 to 6 minutes. Reliable alerts start with reliable data.
PostgreSQL · PL/pgSQL · pgAgent · MyReport BE
Finding: 7 supervised regression tests with no discrepancy in the verified scope.
CLOSED Read the case file →Appendices: forensic glass fragment classification (213 samples, Random Forest, macro F1 0.735) · food price analytics (66,100 WFP records, 2005–2025) · Power BI decision-support prototype (Alshadows, demonstration data) · Hadoop & Spark lab (Docker cluster, MapReduce and PySpark).
Documented findings, linked to the case studies and coursework.
| Skill | D01 | D02 | D03 | D04 | D05 | D06 |
|---|---|---|---|---|---|---|
| Disk and memory forensics | ● | |||||
| Network analysis | ● | ● | ● | |||
| Detection and threshold tuning | ● | ● | ||||
| Understanding lateral movement | ● | |||||
| AI model security | ● | |||||
| Data reliability and traceability | ● | ● |
A workflow suited to each project: collect relevant material, examine it, test hypotheses and report the results. For digital forensics, the approach draws in particular on NIST SP 800-86.
01
Acquire data without altering it and keep a record.
02
Extract what matters: processes, logs, flows, artefacts.
03
Correlate sources and test hypotheses, thresholds included.
04
Write a clear, dated finding that a third party can verify.
Security analysis is only as reliable as the data behind it: trustworthy inputs, effective checks and traceability. At HortiTrace, I applied these principles to production PostgreSQL pipelines; in my cybersecurity projects, they guide how I examine evidence and interpret results.
Built through documented case studies and MSc coursework, with the learning context made explicit for each area.
Experience
Optimised a production PostgreSQL ETL chain: incremental loading and a pre-aggregated reporting layer, then validation through 7 supervised regression tests, with no discrepancy found within the verified scope. Case file D06
Designed a BI prototype on demonstration data: 9-table SQL Server model, Power Query preparation, 20+ DAX measures and 4 Power BI pages.
Data security checks, backups and patching; business requirements analysis, integration of information systems (ERP, business applications), IT project follow-up and user support.
Optimised a TensorFlow vehicle-detection API (accuracy raised from 73% to 88%) and integrated inventory-monitoring dashboards.
Education
Average 16.43/20, ranked 1st in semesters 2 to 4. Detailed results
First-year CGPA: 3.50/4.00.
Degree awarded · CGPA: 3.57/4.00.
Off the record
Adaptation, football, entrepreneurship and a first international experience: a few stories that also explain how I work.
Unrestricted · Contact
