Description

I specialize in building production AI systems including LLM-powered agents, RAG pipelines, and ML infrastructure from prototype to deployment. 5 years of experience shipping AI products with Python, PyTorch, LangChain, and FastAPI, from model optimization to production systems.

At Kog, I've shipped end-to-end AI systems including image generation pipelines, semantic search with vector embeddings, and multi-agent code generation systems with RAG-powered retrieval. My work spans the full stack: from GPU-backed ML endpoints and real-time inference to data pipelines and cross-functional collaboration with product teams.

👨🏻‍💻 Technical Focus:

• GenAI & LLM (Diffusions, Transformers, LoRA, LangChain, LangGraph)

• Production ML infrastructure (FastAPI, PyTorch, Azure, NVIDIA)

• Data pipelines & vector search (PostgreSQL, pgvector, BigQuery)

• Agentic AI & RAG architectures

I'm passionate about creating AI-driven products that empower users. Always open to discussing GenAI, ML infrastructure, and what it takes to ship AI at scale.

Langues

Anglais
Bilingue ou natif
Chinois
Bilingue ou natif

Préférences en matière de lieu de travail

Accepte de travailler sur site

Paris (jusqu’à 50 km)

Kog
Senior AI Product Engineer
janvier 2024 - avril 2026 (2 ans et 3 mois)
• Achieved a 25% reduction in inference latency for a production-grade image generation pipeline by implementing advanced caching, torch compile, and float8 quantization, while integrating complex workflows including ControlNet, IP-Adapter, and LoRA.

• Created AI-powered game creation platform featuring a LangGraph-based coding agent with multi-agent architecture and RAG-powered context retrieval, enabling autonomous game generation and iterative refinement.

• Achieved 8x payload reduction for VLM segmentation masks via custom binary compression, cutting response latency and enabling real-time inference.

• Reduced infrastructure costs by implementing shared model caching across ML servers, eliminating redundant loads and optimizing GPU memory utilization.

• Engineered production ML orchestration servers with Python, FastAPI and Pydantic, partnering closely with the frontend team to integrate NVIDIA A100 GPU-backed ML endpoints into the product with stable, well-specified request/response contracts and Azure cloud storage integration.

• Enabled semantic image search by building vector pipeline (vision embeddings, pgvector), replacing keyword-based retrieval with visual similarity matching.

• Scaled synthetic data generation to 10,000+ samples/day using TypeScript/Puppeteer automation, eliminating manual orchestration across ML servers.

• Accelerated ML development cycles by 2x with Gradio-based QA tooling, enabling rapid visual regression testing and real-time output comparison.

• Iterated on the product by translating UX needs into ML-backed features with strong focus on latency, scalability, and maintainability.
Python RAG AI Agent FastAPI LLM
Arianee
Data Scientist
novembre 2022 - décembre 2023 (1 an et 1 mois)
• Unlocked personalized discovery for users by building an NFT recommendation system with EfficientNet embeddings and hybrid filtering

• Reduced fraud risk by building a Dagster-orchestrated anomaly detection system that automatically flagged suspicious blockchain transactions, with Neo4j visualizations to trace and track fraudulent activity

• Empowered marketing and product teams with user personas by designing Dagster ETL pipelines on GCP that processed blockchain data at scale

• Enabled data-driven operations by architecting a real-time Looker Studio dashboard that monitored 1-2M+ blockchain transactions, integrating BigQuery and PostgreSQL
SaiciAI
Data Engineer
janvier 2021 - janvier 2022 (1 an)
• Engineered automated data collection system using Python and Selenium, extracting structured datasets from 5+ region-specific social media platforms and reducing manual collection time by 90%.

• Designed and deployed scalable ETL pipelines processing 10,000+ records daily, transforming unstructured social media data into ML-ready feature sets in PostgreSQL.

• Integrated Airbyte for real-time data streaming to BigQuery, reducing analyst wait times from hours to minutes and enabling rapid model iteration cycles.

• Built and deployed YOLO-based object detection pipeline for automated content classification across 50+ categories, eliminating manual tagging workflow.