Machine Learning Engineer

I'm a 22 y/o undergrad from Paris, dedicating my time to startups and AI.

Get in touchHugging FaceLinkedIn
Portrait of Amaury Delille

Amaury Delille
48.8566° N, 2.3522° E

Now
Founder, VietBuddyFounder, Migraine0Machine Learning Intern, Groupe Bouygues
Before
Machine Learning Intern, Bosch
Focus
Embodied AI & consumer apps
Studying
MSc Computer Science & AI, EPITA
Based
Paris, France
01/Experience6 roles · 2023 → now

Where I've been shipping.

  1. Three VietBuddy App Store screenshots: a speaking drill, a voice conversation with Bé Trâu, and a flashcard word list

    Fig. 01VietBuddy on the App Store

    01/Aug 2026 — NowOngoing

    VietBuddy

    Founder · Remote

    iOS Vietnamese learning app designed to help learners speak real, everyday Vietnamese with confidence.

    • Voice conversations with Bé Trâu, an AI speaking partner, instead of flashcard drills.
    • Scaling to $1M ARR.
    • iOS
    • Voice AI
    • Consumer
    • Growth
  2. Three Migraine0 App Store screenshots: daily diary, recurring patterns, 28-day overview

    Fig. 02Migraine0 on the App Store

    02/Jul 2026 — NowFor saleOngoing

    Migraine0

    Founder · Remote

    Your migraine tracker, without the daily burden. Confirm each day in seconds, uncover possible patterns, and prepare a clear 28-day record for your next appointment.

    • 100k+ views on social media.
    • Live on the App Store with paying subscribers since July 2026.
    • iOS
    • Health
    • Pattern Mining
    • Consumer
    Migraine0 is for sale. Interested? Email me
  3. AgiBot humanoid robot standing in a physics simulation with lidar points scattered across the floor

    Fig. 03Humanoid in simulation, lidar sweep visible

    03/Jun 2026 — NowOngoing

    Groupe Bouygues

    Machine Learning Intern · Paris, France

    Embodied AI on a humanoid robot: manipulation, perception and on-device multimodal models.

    • Developing AI capabilities for the Agibot A2 Ultra humanoid robot ahead of its VivaTech 2027 showcase.
    • Trained SAC policy for OmniHand object handling.
    • Deployed an RT-DETR object detection pipeline for real-time customer-intent classification and robot interaction triggering, averaging 81ms latency.
    • Deploying on-device VLMs (Qwen models) for low-latency multimodal inference.
    • Reinforcement Learning
    • SAC
    • RT-DETR
    • VLMs
    • Robotics
  4. Ho Chi Minh City skyline at night with Bitexco Financial Tower lit up

    Fig. 04Ho Chi Minh City at nightPhoto: Jim 陳, CC BY-SA 2.0

    04/Sep 2025 — Feb 20266 mos

    Bosch

    Machine Learning Intern · Ho Chi Minh City, Vietnam

    Time-series forecasting, retrieval and knowledge graphs for Bosch Automobile.

    • Implemented and trained decoder-only next token prediction models for time-series forecasting tasks.
    • Fine-tuned time-series forecasting transformers using LoRA for anomaly detection.
    • Developed a RAG-powered search platform with 250ms average latency, streamlining enterprise knowledge retrieval across Bosch Automobile, based on an ANN for retrieving and a cross-encoder + BM25s pipeline for re-ranking.
    • Built an AI-agent-based ETL framework, converting Excel sheets into Neo4j-ingestible graph schemas, removing tedious manual data onboarding and enabling knowledge graph ingestion.
    • PyTorch
    • Time Series
    • LoRA
    • RAG
    • Neo4j
  5. Long underground wine cellar corridor lined with racks of bottles

    Fig. 05Underground cellar, Château Pontet-CanetPhoto: seligmanwaite, CC BY 2.0

    05/Dec 2024 — Jun 20257 mos

    Wine Labs

    AI Engineer, Freelance · New York, United States

    Data infrastructure for the wine trade, remote from Paris.

    • AI agent scraper deployed on 2M+ websites.
    • Outlier detection using statistical methods.
    • Optimized real-time backend events and data communication.
    • AI Agents
    • Scraping
    • Outlier Detection
    • Real-time
  6. Liège Airport passenger terminal under a clear blue sky

    Fig. 06Liège Airport passenger terminalPhoto: Raimond Spekking, CC BY-SA 4.0

    06/May 2023 — Jul 20233 mos

    Pyxis Technologies

    Software Engineer Intern · Brussels, Belgium

    Client software for Liège Airport and in-house web tooling.

    • Software development, bug and issue solving for Liège Airport using C#/.NET, RabbitMQ, Entity Framework and SQL.
    • WordPress plugin development with PHP.
    • Agile training.
    • C#
    • .NET
    • RabbitMQ
    • SQL
    • PHP
02/Projects5 entries

Things I built on the side.

P01/In stealth

Stealth Startup

Something is brewing.

I'm exploring fragmented markets that are ripe for AI, industries where business intelligence and aggregation are still done with spreadsheets, phone calls and gut feeling.

  • Business Intelligence
  • Aggregation
  • AI
P02/Archived

Condense Labs

Compress LLM models. Deploy anywhere.

LLMs are huge. We made them faster, lighter and smarter, compression pipelines built on knowledge distillation, quantization and pruning, with impressive results along the way.

  • Distillation
  • Quantization
  • Pruning
P03/8B params

Tana-8B

A decoder-only LLM trained on a French corpus.

Built so my computer-science-informed grandpa, a heavy Copilot and ChatGPT user, could finally relate to what I do. RoPE encoding, SwiGLU MoE and FlashAttention, trained distributed across two GPUs with ZeRO-3. Stopped at 1/10 of the dataset because it was getting too expensive…

  • RoPE
  • SwiGLU MoE
  • FlashAttention
  • ZeRO-3
P04/On Hugging Face

PhoGPT-0.13B

English–Vietnamese translator, written by hand in PyTorch.

Born from a boring afternoon — no business purpose, pure pleasure. It struggles with complex sentences given the model (130M params) and dataset size (1M rows), but writing everything by hand taught me more about LLMs than any course.

  • PyTorch
  • Transformers
  • Translation
P05/250+ users

Sanily Skincare

Know what you consume.

I break down skincare ingredients so you don't have to. My first serious project, born from frustration with apps full of commercial bias, outdated data and zero fact-checking. 250+ users across Indonesia, France, the US and more. Earned $60 in subscriptions, lost ~$150 to infra and App Store costs, and learned a fortune in system design, ML pipelines and startup reality.

  • System Design
  • ML Pipelines
  • iOS
03/Notes1 entry

Field notes from the road.

04/Record

Hackathons & school.

Education

  • EPITA

    2022 — 2027

    MSc Computer Science & AI · Paris

    Five-year master's in Computer Science and AI. First three years: low-level programming (OS, C, C++, ASM), mathematics (linear algebra, calculus, probability & statistics) and data structures & algorithms. Last two years: ML algorithms (neural nets, RL, Bayesian inference), advanced mathematics (convex optimization, lin alg II, proba & stats II) and applied ML (recommender systems). Enrolled in the Entrepreneur Program since 2024, presenting and iterating on a startup project.

  • FPT University

    2024

    Exchange Program — OS, ML & SWE · Vietnam

    A life-changing experience. I discovered a new culture, new people, new ways to interact and make friends, and truly fell in love with Vietnamese culture.