Tushar Kumar

Tushar Kumar

Nov 9

The Compiler Trusts the Graph: An Adversarial Evaluation of a Toy ML Compiler

#ai #security #llm #machinelearning

11 min read

Tushar Kumar

Nov 8

Close Enough to Be Wrong: An Adversarial Evaluation of a Semantic Cache

#ai #security #llm #machinelearning

10 min read

Tushar Kumar

Nov 7

The Judge Reads Tokens: An Adversarial Evaluation of an Eval Toolkit's Own Scorers

#ai #security #llm #machinelearning

11 min read

Tushar Kumar

Nov 6

Citation Is Not Truth: A Three-Metric Evaluation of an Extractive Research Pipeline

#ai #security #llm #machinelearning

1

7 min read

Tushar Kumar

Nov 5

The Gate Is a Vocabulary: A Three-Metric Evaluation of a Voice Agent's Confirmation Step

#ai #security #llm #machinelearning

11 min read

Tushar Kumar

Nov 4

The Gate Held, the Draft Leaked: A Three-Metric Evaluation of a Support Agent's Safety Layers

#ai #security #llm #machinelearning

1

8 min read

Tushar Kumar

Nov 3

Retrieval Is Not a Safety Boundary: Poisoned-Corpus and Unsupported-Question Evaluation of a RAG Pipeline

#ai #llm #security #machinelearning

1

7 min read

Tushar Kumar

Nov 3

Pattern Guardrails Under Mutation: A Three-Metric Evaluation of Two Agent Safety Layers

#ai #llm #security #machinelearning

6 min read

DEV Community

Badges

Writing Debut

The Compiler Trusts the Graph: An Adversarial Evaluation of a Toy ML Compiler

Close Enough to Be Wrong: An Adversarial Evaluation of a Semantic Cache

The Judge Reads Tokens: An Adversarial Evaluation of an Eval Toolkit's Own Scorers

Citation Is Not Truth: A Three-Metric Evaluation of an Extractive Research Pipeline

The Gate Is a Vocabulary: A Three-Metric Evaluation of a Voice Agent's Confirmation Step

The Gate Held, the Draft Leaked: A Three-Metric Evaluation of a Support Agent's Safety Layers

Retrieval Is Not a Safety Boundary: Poisoned-Corpus and Unsupported-Question Evaluation of a RAG Pipeline

Pattern Guardrails Under Mutation: A Three-Metric Evaluation of Two Agent Safety Layers