Loading
Ace Intelligence Systems
Preparing an ultra-fast, streamlined experience.
Loading
Preparing an ultra-fast, streamlined experience.
Ace Intelligence
Turn hours of manual security log triage and threat analysis into verified defensive playbooks in seconds.
When a potential security alert triggers, junior analysts and developers often struggle to quickly correlate complex MITRE ATT&CK patterns, server log anomalies, and cloud misconfigurations. PralayAI is a purpose-fine-tuned cybersecurity assistant engineered strictly for defensive triage, threat mitigation, and compliance guidance.
Security teams face acute alert fatigue, sifting through hundreds of raw server logs, firewall alerts, and vulnerability scanner reports every day. Mid-sized companies lack 24/7 senior security analysts, leading to delayed response windows during active breach attempts and compliance audits.
A fine-tuned open-source defensive model trained on thousands of security playbooks, incident response protocols, and cloud hardening frameworks. It analyzes raw logs, explains attack vectors, and outputs step-by-step mitigation playbooks without exposing sensitive internal logs to public cloud LLM APIs.
Rapid Incident Triage: Translates confusing firewall and auth logs into clear, actionable incident timelines. | MITRE ATT&CK Mapping: Instantly identifies attacker tactics and suggests specific defensive countermeasures. | Air-Gapped Private Deployment: Fine-tuned open-source model can run completely on-premises or on private CUDA infrastructure with zero corporate data leakage. | Strict Guardrails: Hardened to refuse offensive weaponization, malware generation, and phishing crafting.
PralayAI enforces a strict defensive-only policy. The model is hardened against adversarial prompts and refuses requests for weaponized exploits, malware, and phishing. Automated evaluation notebooks run 8 defensive scenarios and 5 adversarial penetration tests with verified refusal fidelity.
Defensive Scope:
Incident Response | Log Analysis | Threat Detection
MITRE ATT&CK Mapping | Cloud Security | Hardening
Blocked Exploitative Topics:
Phishing | Credential Theft | Malware | Ransomware | EvasionBuilt on Qwen2.5 1.5B Instruct, fine-tuned with QLoRA using the Unsloth framework on a curated cybersecurity dataset. Achieves ~4.5-second inference latency on dedicated local CUDA GPUs with full conversational persistence.
Base Model: Qwen/Qwen2.5-1.5B-Instruct
Fine-Tuning: Unsloth + QLoRA
Inference Latency: ~4.5s on GPU
Model Artifacts: Hugging Face OMCHOKSI108/Paralay1.1-MergedFour-tier architecture: React dashboard frontend communicates with a FastAPI backend, which stores conversation history in PostgreSQL and routes inference requests to either local CUDA GPU workers or cloud endpoints. Designed for private VPC or air-gapped deployment.
Python | FastAPI | React | PostgreSQL | Qwen2.5 | QLoRA | Unsloth | Hugging Face | DockerExplore the complete open-source codebase, fine-tuning notebooks, merged model weights, and evaluation harness on GitHub.
https://github.com/OMCHOKSI108/pralayAIArchitected and delivered by the Ace Intelligence engineering team.
Om Choksi (CTO) — https://github.com/OMCHOKSI108We design, engineer, and deploy production-grade software and AI systems that eliminate manual bottlenecks, reduce operating costs, and accelerate revenue.