AI & Machine Learning Engineer — Gwalior, India
Building systems that learn — from RLHF-based LLM training to production-grade computer vision and predictive pipelines.
I'm a Data Science & Machine Learning professional with hands-on experience in RLHF-based LLM training, deep learning architectures (CNN, LSTM), and building end-to-end ML pipelines from messy raw data to deployed, production-ready systems.
At Ethara.AI, I evaluated large-scale conversational datasets for LLM improvement — sustaining 95%+ annotation accuracy across six quality dimensions including safety, coherence, and hallucination detection across thousands of samples.
My project work includes a biometric authentication system that hit 97% face recognition accuracy on 24,000+ training images, and a multi-module agricultural intelligence system processing over 10,000 crop records for price forecasting and trend analysis.
Currently pursuing B.E. Computer Science at MITS Gwalior (2024–27), actively seeking AI/ML engineering roles where I can build systems that matter.
# Subhash Dangi class MLEngineer: name = "Subhash Dangi" location = "Gwalior, India" status = "Open to work" expertise = [ "RLHF / LLM Training", "CNN / LSTM", "Computer Vision", "Predictive Modeling", "NLP & Data Annotation", ] education = [ "B.E. CSE — MITS (2024–27)", "Diploma — SRGP (2021–24)", ] def goal(self): return "Build systems that matter."
A real-time AI translation app deployed live on Hugging Face Spaces. Demonstrates end-to-end deployment of language models — users can translate text instantly through a clean web interface without any setup.
A 3-module predictive system for crop price forecasting, market trend analysis, and seed quality assessment. Processed 10,000+ records end-to-end, improving input quality by ~30%. Models evaluated on Accuracy, Precision, Recall, F1, and RMSE.
Multimodal CNN-based system for face and fingerprint recognition. Trained on 24,000+ face images across 12 classes and 2,000+ fingerprint samples. Achieved 97% face recognition accuracy — production-level reliability confirmed via FAR, FRR, and Confusion Matrix analysis.
Open to full-time roles, internships, and collaborations.
If you have something interesting — let's talk.