AI Evaluation & Benchmark Engineer focused on LLM Evals, Agent Reliability, and Post-Training.
I build evaluation harnesses, benchmarks, and test datasets to understand where AI systems fail — across reasoning, retrieval, tool use, structured outputs, and multimodal tasks.
My work spans benchmark design, golden datasets, deterministic & model-based evaluation, failure analysis, regression testing, and AI system observability.
I also build Agentic AI and RAG systems, which helps me evaluate them from both sides: how they are built and how they break.
- LLM & Agent Evaluation
- Benchmark Design
- Golden Dataset Construction
- Agent & Tool-Use Reliability
- Multimodal Evaluation
- Failure Analysis & Regression Testing
- RAG Evaluation
- Post-Training Evaluation
- Long-Horizon Agent Evals
A reproducible multimodal benchmark for evaluating vision-language models on fine-grained visual reasoning.
- Provider-agnostic evaluation harness
- Structured JSON outputs
- Deterministic one-to-one scoring
- Precision, recall, attribute accuracy & coverage
- Multi-model benchmarking
- Regression tests for evaluation correctness
→ github.com/kinba09/spotdiff-eval
I write about the engineering behind evaluating AI systems:
- LLM & Agent Evals
- Benchmark Design
- Agent Failure Modes
- RAG & Tool-Use Reliability
- AI Observability
- Production AI Systems
Blog: kinba09.github.io/blogs.html
Evals & Benchmarks
LLM Evals • Benchmark Design • Golden Datasets • Deterministic Scoring • LLM-as-a-Judge • Regression Testing • Failure Analysis • Safety Testing
AI Systems
Agentic AI • RAG • LangGraph • LangChain • LlamaIndex • MCP • Tool Calling • Structured Outputs
Engineering & Observability
Python • FastAPI • SQL • Docker • OpenTelemetry • Prometheus • CI/CD
- LinkedIn: linkedin.com/in/abnik-ahilasamy
- X: x.com/Abnik_Ahilasamy
- Website: kinba09.github.io
- Blog: kinba09.github.io/blogs.html
- Email: abnikahilasamy@gmail.com