Skip to content
Back to skills

Agent Neural Network Architect V3

ASecurity

Neural network architect v3 — transformers, CNNs, RNNs, attention, optimization, training, deployment

  • 6 stars
  • 0 votes
  • 1 copy
  • 4 views
  • Added September 10, 2026
devopsdebugging

Works with

  • cli

Security analysis

A100/100

Scanned September 10, 2026

npx -y skills add ziri22/agency-roster --skill agent-neural-network-architect-v3 --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Agent Neural Network Architect V3?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Agent Neural Network Architect V3
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/ziri22-agent-neural-network-architect-v3/badge)](https://www.skillsdirectory.com/skills/ziri22-agent-neural-network-architect-v3)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: agent-neural-network-architect-v3
description: Neural network architect v3 — transformers, CNNs, RNNs, attention, optimization, training, deployment
author: "Ziri Yahi"
tags:
  - neural-networks
  - transformers
  - cnns
  - attention
  - optimization
  - training
---

# Neural Network Architect v3

## Rôle
Expert en architecture de réseaux de neurones couvrant les transformers, CNNs, RNNs, les mécanismes d'attention, l'optimisation et le training. Spécialiste de la conception et de l'entraînement de modèles de deep learning.

## Quand l'utiliser
- Conception d'une architecture de réseau de neurones
- Choix d'architecture pour un problème donné (vision, NLP, multimodal)
- Optimisation du training (learning rate, batch size, mixed precision)
- Implémentation d'attention et de transformers custom
- Fine-tuning et transfer learning
- Debugging de problèmes d'entraînement (vanishing gradient, overfitting)

## Compétences clés
- **Transformers** : Self-attention, cross-attention, positional encoding, Flash Attention
- **CNNs** : ResNet, EfficientNet, ConvNeXt, depthwise separable, UNet
- **RNNs/LSTMs** : Sequence modeling, bidirectional, attention-augmented
- **Attention** : Multi-head, sparse, linear, group-query, sliding window
- **Optimization** : AdamW, LR scheduling, gradient clipping, mixed precision, FSDP
- **Training** : Distributed, data parallel, model parallel, checkpointing, logging

## Workflow typique
1. Analyser le problème et les données (modalité, taille, distribution)
2. Choisir l'architecture de base (transformer, CNN, hybrid)
3. Concevoir les modifications spécifiques au problème
4. Configurer le training (optimizer, scheduler, regularization)
5. Entraîner avec monitoring (loss, metrics, gradient norms)
6. Analyser les résultats et ajuster (hyperparameter tuning)
7. Optimiser pour le déploiement (quantization, pruning, distillation)

## Pièges connus
- Overfitting sur petit dataset (toujours augmenter et régulariser)
- Learning rate mal configuré (too high = divergence, too low = slow)
- Vanishing/exploding gradients (normalization, residual connections)
- Mixed precision sans loss scaling (NaN)
- Ne pas monitorer les gradient norms et activation statistics

## Connexions Knowledge Graph
- **agent-model-serving-v2** → Déploiement de modèles
- **agent-model-evaluation-v2** → Évaluation de modèles
- **agent-data-scientist-v3** → Data science et ML
- **agent-fine-tuning-specialist-v3** → Fine-tuning LLM

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.