Looking for a job

sushain devi

@sushaindevi

Message

I research geometric interventions that improve truthfulness and reasoning in quantized LLMs.

India

Message

What I'm looking for

I'm looking to advance research and engineering for trustworthy LLMs, mechanistic interpretability, causal interventions, and reasoning systems, where I can turn geometric theory into practical methods for reducing hallucinations and improving self-correction.

I'm the author of Project NIKA, a five-paper research series on truth circuits, hallucination correction, and epistemic agency in quantized LLMs. Using consumer-grade Kaggle and Colab T4 GPUs, I've developed neuro-symbolic, causal, geometric, and gauge-theoretic frameworks for understanding model reasoning failures.

I've also built LLM observability, legal-document NLP, spam detection, semantic retrieval, document intelligence, and recommendation systems through roles at KloudMate, Entertainment Technologist, and Afame Technology. My work spans PyTorch, Hugging Face, LoRA fine-tuning, mechanistic interpretability, and real-time neural-circuit monitoring.

Experience

Work history, roles, and key accomplishments

Associate Developer Intern

KloudMate

Jun 2025 - Dec 2025 (6 months)

Architected AI-powered observability platform integrating LLM-driven Copilot, NLU pipeline, and OpenLLMetry dashboards for intelligent system monitoring. Developed intent classification and entity extraction systems; implemented agent-based tool routing for logs, metrics, and traces.

LLM NLU OpenLLMetry Intent Classification Entity Extraction AI Agents Telemetry APIs Anomaly Detection

AI Developer Intern

Entertainment Technologist

Jul 2024 - Feb 2025 (7 months)

Engineered end-to-end NLP pipeline for automated legal document processing in film industry. Implemented text extraction, key phrase identification, and abstractive summarization on domain-specific documents.

NLP Text Extraction Keyword Extraction Abstractive Text Summarization Document Processing

Machine Learning Intern

Afame Technology

Apr 2024 - May 2024 (1 month)

Refined SMS spam detection algorithms through collaborative model optimization. Achieved 15% accuracy improvement through feature engineering and ensemble methods.

Machine Learning Feature Engineering Ensemble Methods Spam Detection