Amaljith Kuttamath
Retrieval, evaluation,
and interpretability.
I build data agents and retrieval systems, and study how language models represent information. My projects connect practical engineering with experiments in model behavior.
Research projects
Research libraries & replications
Five open research projects. Choose a question to follow the relevant work.
trust-bench
Interpretability toolkit probing SAE features, activations, and circuits in Llama 3.1 8B.
jlens-replication
Replicating Anthropic's Jacobian lens on small open-weight models with free compute.
sae-explorer
Investigating a Gemma 2 2B SAE feature associated with conjunctions across six languages.
mlx-interp
Mechanistic interpretability tools for Apple MLX. Sparse autoencoders on Apple Silicon.
hybrid-attention-mlx
From-scratch MLX implementation of Qwen3.5 hybrid DeltaNet + full-attention architecture.
Background
I started at ISRO, then co-founded RadianArc, building digital twins for predictive maintenance on industrial machinery.
I later worked on retrieval systems and knowledge graphs for healthcare.
I completed an MS in Data Analytics at George Mason University, with a thesis on RAG systems for healthcare.
Read my résuméEducation and publications
- Education
-
MS, Data Analytics
George Mason University - Certification
-
Google Cloud Professional
ML Engineer, 2024 - Publication
-
Predicting Risk of Lung Cancer ↗
Quality Management in Health Care
Work with me
I work on retrieval, evaluation, and model-backed products. We start with one concrete problem, agree on what success looks like, then build a working slice and test it.
Project and research enquiries:
kuttamath.amaljith@gmail.com