NLP · RAG · LLM AGENTS · GRAPH ML

AI Engineer building RAG & LLM agents measured, not guessed.

Final-year engineering student at EPITA (Data Science & AI, Graphs track). At Merck Group I built and benchmarked a RAG assistant and a pull request review agent. I am looking for a 6-month end-of-studies internship from February 2027.

View projects Contact
3+ Years across AI & data
3 Countries: Cameroon, Senegal, France
14 Certifications
Portrait of Aboubakar Saly

Focus

NLP • RAG • Graph ML

Availability

6-month internship • Feb 2027

Current Base

Lyon, FR

01 — Profile

Details & Context

I build AI tools that answer from real data, and I measure whether they work.

Final-year engineering student at EPITA in Lyon, majoring in Data Science & AI on the Graphs track. At Merck Group I ranked candidate AI use cases, then built the selected one: a RAG assistant over code, tickets and documentation, extended to automated pull request reviews. Before EPITA I studied AI & Big Data at UCAD and Telecommunications at ESMT in Dakar, and built Next.js sites for accounting firms in West Africa.

Latest role

Merck Group

AI Systems Architect Intern, Sep 2025 to Jan 2026

Academics

EPITA — SCIA-G Major

Data Science & AI, Graphs track, class of 2027

Focus areas

RAG assistants with measurable qualityLLM agents with human-in-the-loop controlsNLP, knowledge graphs & Graph ML
Discord CV (EN) CV (FR)

Stack DNA

tools

Generative AI

focus
RAG, embeddings & ChromaDB 90%
LLM agents: LangGraph, Claude Agent SDK, MCP 85%
LLM evaluation: RAGAS, benchmarks 85%

Machine Learning

focus
NLP: Hugging Face Transformers 85%
PyTorch & scikit-learn 82%
Graph ML & knowledge graphs: PyKEEN, SPARQL 78%

Data

focus
Python: pandas, NumPy 92%
SQL & DuckDB 85%
PySpark & Big Data 72%

Engineering

focus
FastAPI & REST APIs 85%
Docker, CI/CD & Git 84%
TypeScript, Next.js & Astro 76%

03 — Experience

Internship 01

AI Systems Architect Intern

Merck Group • Molsheim, France (hybrid)

Sep 2025 - Jan 2026

Enterprise Architecture team, Merck Life Science. Prioritized candidate AI use cases, then built and benchmarked the selected one: an assistant for developers working on a large internal codebase.

  • Ranked candidate AI solutions on weighted criteria (feasibility, fit, time-to-value, risk), with a SWOT per option, a PoC decision matrix and an executive summary
  • Built a RAG assistant that answers questions on internal code, tickets and documentation and cites its sources
  • Automated pull request reviews by combining Bitbucket, Jira, SonarQube and Jenkins context into structured feedback for developers
  • Compared four chunking strategies and evaluated answers with benchmark questions, RAGAS metrics and user feedback: answer relevancy 0.93 vs 0.80 for the baseline tool
  • Built a Bitbucket MCP server and a UI branding review agent with the Claude Agent SDK; packaged with Docker and automated with n8n
PythonRAGChromaDBRAGASOpenAI & Claude APIsClaude Agent SDKMCPGradioDockern8n
Freelance 02

Freelance Web Developer & Site Admin

Mac Consulting Senegal • Dakar, Senegal, then remote

May 2024 - Aug 2025

Websites for accounting firms in West Africa, delivered alongside my studies.

  • Built, secured and maintained Next.js sites for accounting firms, hosted on OVH
  • Handled hosting and incidents, then trained client teams to run content and routine operations
Next.jsReactTypeScriptOVHSecurity
Community 03

Forum Member

OpenAI Community • Remote

Oct 2024 - Present

Active contributor sharing experiments and product feedback with the OpenAI builder network.

  • Synthesized suggestions for new tooling & safety flows
  • Partnered with community members on prototypes
  • Documented best practices for new forum members
AI AdvocacyWritingCommunity
Volunteer 04

Association Member

Eau pour Tous Cameroun • Douala, Cameroon

Sep 2022 - Present

Non-profit focused on clean water access in Cameroon's Far North region.

  • Organized fundraising campaigns with 85% higher participation
  • Coordinated logistics and communications for community events
  • Strengthened member engagement through weekly cadences
OperationsProject ManagementCommunity
Internship 05

IT Intern

SODECOTON • Douala, Cameroon

Aug 2022 - Oct 2022

Kept production software online across departments and supported expansion work.

  • Maintained business applications for cross-team reliability
  • Documented troubleshooting playbooks for the service desk
  • Strengthened backups and monitoring for remote offices
SystemsMonitoringSupport

02 — Projects

Follow builds
Featured 01

RAG Assistant & PR Review Agent

Merck Group internship, 2025-2026

Assistant that answers developer questions from internal code, tickets and documentation with cited sources, extended to automated pull request reviews that combine Bitbucket, Jira, SonarQube and Jenkins context.

PythonRAGChromaDBRAGASGradioDockern8n
Chunking strategies compared 4
Sources Code, tickets, docs
Confidential, internship work
Featured 02

TradeMind: Governed LLM Trading Agent

EPITA Agentic AI course, 2026

LLM agent that proposes buy, sell or hold decisions on five assets in paper trading only. A critic agent, deterministic risk limits and mandatory human approval gate every simulated order. No real-money endpoint exists.

LangGraphLangChainFastAPIStreamlitSQLitePydantic
Automated tests 103
Human approval Every order
Course project, code on request
Featured 03

Horizon Care: Clinical Information Extraction

EPITA end-of-studies group project, 2026

Privacy-first assistant for medical consultations. It watches the practitioner screen, extracts medications and diagnoses, and benchmarks a local OCR pipeline against a cloud vision-language model on accuracy, latency and cost.

PythonEasyOCROpenCVVision LLMDuckDBOpenEMR
F1, medications & diagnoses 1.0
Recommendation Local, private
Team project, school GitLab
Featured 04

Multi-label Toxic Comment Classification

NLP project, Jigsaw dataset

Compared five approaches on six toxicity labels, from TF-IDF with linear models to BiLSTM and fine-tuned DistilBERT. Tuned per-label thresholds and analysed errors on rare classes.

PyTorchHugging FaceDistilBERTscikit-learnBiLSTM
Approaches compared 5
Labels 6
View code
Featured 05

Scientific Collaboration Knowledge Graph

EPITA Graphs track, lead contributor

RDF/OWL knowledge graph of a research lab built from DBLP. Community detection compared against the real teams, and retrospective link prediction to suggest future collaborations.

RDF/OWLSPARQLNetworkXLeidenPyKEEN
Authors / co-publications 786 / 2,659
Community NMI 0.71
Team project, school GitLab

More projects

11 projects

DataValueXplo

Team project

EPITA team project, 2026

Ranks traffic-disrupting accidents so a local authority knows which zones and time slots to study first. PySpark ML on 1.74M rows, with a Streamlit dashboard.

PySparkSpark MLStreamlitpytest

PyAllumette

Team project

EPITA team project, 2026

Hand-written autodiff and MLP driving a simulated race car with model predictive control, compared with a learned-dynamics MPC.

PythonNumPyAutodiffMPC

CarbonLens

Hackathon team

Green IT hackathon, 2026

Compares the estimated carbon footprint, cost and latency of cloud architectures across providers, with uncertainty ranges, using Boavizta data.

Node.jsBoaviztaGreen IT

Bitbucket MCP Server

Private repo

Developer tooling

MCP server for self-hosted Bitbucket Server: repository browsing, pull request workflows, diffs and comments for LLM agents.

TypeScriptMCPLLM agents

Local Audio Transcription

Personal tool

Personal tool

Offline transcription of long audio and video files to .txt, .srt and .vtt with faster-whisper large-v3 on a consumer GPU.

Pythonfaster-whisperCUDA

Malware Classifier

Code

Machine learning

Malware detection with several models and automated hyperparameter optimization workflows.

PythonLightGBMFastAPI

Accounting Web Platform

Code

Full stack

TypeScript accounting platform built around real business workflows for firms in West Africa.

TypeScriptWeb AppAccounting

School Management System

Code

Team project

Modular Java desktop suite for enrollment, scheduling, exams and finance.

JavaJavaFXMySQL

Parallel Computing Toolkit

Code

Algorithms

Java algorithms exploring parallel and distributed computation patterns for intensive workloads.

JavaParallelism

IoT Smart Lamp Controller

Code

IoT

Connected lamp prototype with a lightweight web interface for remote control.

IoTWeb UIHardware

Web Scraping Toolkit

Code

Automation

JavaScript toolkit that automates data extraction from structured websites.

JavaScriptScrapingAutomation

04 — Certifications

14 credentials
AI & Machine Learning #1

CS50's Introduction to Artificial Intelligence with Python

Harvard University

Jun 2025

AI & Machine Learning #2

Microsoft Certified: Azure AI Fundamentals

Microsoft

Jun 2025

AI & Machine Learning #3

Artificial Intelligence Fundamentals

IBM

May 2025

Web Development #4

Introduction to Front-End Web Development

Meta

May 2025

Professional #5

Prévention Sup' - Santé et Sécurité au Travail

INRS France

May 2025

AI & Machine Learning #6

Découvrir l'IA Générative

LinkedIn

Apr 2025

Programming #7

Foundational C# with Microsoft

freeCodeCamp

Apr 2025

DevOps #8

GitHub Foundations

GitHub

Apr 2025

Data Science #9

Introduction to Data Science

Cisco Networking Academy

Apr 2025

Web Development #10

JavaScript Essentials

Cisco Networking Academy

Apr 2025

Finance #11

Les Fondements de la Fintech

LinkedIn

Apr 2025

Languages #12

CEFR Level C1 English

Rosetta Stone

Jan 2025

Languages #13

TOEIC: 865/990

ETS

Jan 2025

DevOps #14

Docker

TestDome

Mar 2024

14 Total Certifications
12 Unique Issuers
2025 Latest Achievement

05 — Contact

Let's collaborate

Looking for an AI engineer who ships RAG systems and LLM agents with measured results?

I am available for a 6-month end-of-studies internship from February 2027, in France or abroad. Tell me about your team and the problem you want to solve.

Compose email

Email

aboubakarsaly@aol.com

Location

Lyon, France

Ask me about Aboubakar's RAG work at Merck.

AI assistant

Prométhéon

Hi! I'm Prométhéon, Aboubakar's AI assistant. Ask me anything about his experience, skills, or availability. (Je parle aussi français et d'autres langues!)
Model: AI Assistant