Welcome to AKIRA

Automatic Test Plan Generator for Hardware Certification

Powered by Transformer & Gemini-based AI Models

What is AKIRA?

AKIRA is an intelligent system that automates test plan generation for hardware certification. It analyzes device specifications and hardware components to produce comprehensive, JSON-formatted test plans that streamline certification workflows.

The system combines a fine-tuned DistilBERT transformer model for component classification with Google Gemini for structured data extraction from vendor datasheets. A Retrieval-Augmented Generation (RAG) pipeline backed by ChromaDB further enhances prediction accuracy by grounding results in real certification data.

What does it do?

Given a hardware component name (like "Intel I350 Gigabit Network" or "NVIDIA T1000 GPU"), the system predicts:

  • Component Type: GPU, Network, Storage etc.
  • Test Name: The certification test suite applicable to this component
  • Features: Specific capabilities like "25 Gigabit Ethernet" or "Accelerated GPU Graphics"
  • Classification Label: High-level category for organizing components

How does it work?

1

Input Component Name

Enter the hardware component name or description

2

Transformer + Gemini Analysis

DistilBERT classifies the component; Gemini extracts structured data from datasheets

3

Generate Predictions

Receive structured JSON output with all attributes

4

Use Results

Copy JSON output for integration with your systems

Single

Single Predict

Get instant predictions for a single hardware component with detailed confidence scores.

Try Single Predict →
Batch

Batch Predict

Process multiple components at once. Perfect for analyzing entire system configurations.

Try Batch Predict →
Spec

Datasheet Prediction

Extract components from vendor spec sheets and PDFs using Gemini + RAG for automatic test plans.

Analyze Datasheet →
Cloud

Cloud Instance

Generate test plans from AWS and Azure VM documentation URLs automatically.

Generate Test Plan →

Key Features

•
Transformer-Powered

Fine-tuned DistilBERT model trained on 21,000+ hardware certification records

•
Gemini Extraction

Google Gemini parses vendor datasheets, PDFs, and spec pages into structured components

•
RAG Pipeline

ChromaDB-backed retrieval augments predictions with real certification data

•
Multi-Vendor Support

Works with Dell, Lenovo, HPE, Cisco, Supermicro, Fujitsu, and 20+ vendors

•
JSON Output

Structured format ready for direct integration with certification systems

•
Production Ready

API authentication, rate limiting, SSRF protection, and structured logging built in