Looking for a job

Sudhanshu Shandilya

@sudhanshushandilya

Message

AI/ML engineer at Privafy who built an anomaly detection pipeline processing 60M+ events per week per endpoint.

India

Message

At Privafy, I built an end-to-end pipeline for behavioral anomaly detection, processing 60M+ events per week per endpoint. I also developed and evaluated Transformer- and GNN-based models for anomaly and malicious-activity detection, improving on the existing baseline in internal testing.

At Fortinet, I built a code-intelligence and RAG system and fine-tuned Llama and Qwen models for telemetry analysis, reducing triage time by 40% in a controlled evaluation. I also developed DWHP, a multi-seasonal forecasting algorithm that reduced false alerts by 60% and achieved 33% lower MAE than Prophet in benchmarking.

Experience

Work history, roles, and key accomplishments

Senior AI/ML Developer

Privafy

Mar 2026 - Sep 2026 (6 months)

Built an end-to-end AI/ML pipeline for large-scale behavioral anomaly detection, processing 60M+ events/week per endpoint. Developed and evaluated Transformer- and GNN-based deep learning models for anomaly and malicious-activity detection.

Python PyTorch Transformers GNN Anomaly Detection ML Pipeline Design Large Scale Data Processing

Data Scientist 2

Fortinet

Sep 2024 - Feb 2026 (1 year 5 months)

Built a multi-language code-intelligence and RAG system using Tree-sitter, Qdrant and Memgraph. Fine-tuned Llama and Qwen models up to 30B parameters using SFT, LoRA, QLoRA and PEFT, reducing triage time by 40% on a controlled evaluation dataset.

Python PyTorch RAG LLM Fine tuning SFT LoRAs QLoRA PEFT Tree Sitter Qdrant Memgraph Model Quantization Caching Batch Processing Indexing Reranking Pruning Time Series Anomaly Detection

Data Scientist

Fortinet

Jun 2022 - Sep 2024 (2 years 3 months)

Designed and deployed DWHP (Dynamic Weighted Hourly Forecast), replacing static thresholds with adaptive multi-seasonal forecasting and reducing false alerts by 60%. Developed ML and statistical-analysis pipelines for large-scale network telemetry, processing 200K+ records/week.

Python Time series Forecasting Anomaly Detection Statistical Analysis Predictive Analytics Large Scale Data Processing