Nijesh Kanjinghat
AI Engineering Lead, APAC @ IBM
Engineering reliable, scalable, and aligned AI systems
I build agent harnesses, control planes, and multi-agentic systems — with alignment, evaluation, and robustness at the core — spanning inference optimization, evaluation frameworks, and governance for scalable environments.
WHAT I DO
LLM Systems & Performance
MLOps/LLMOps pipelines, LLM inference optimization, and hardware-aware performance engineering on GPUs and TPUs using Triton and Pallas.
Evaluation & Agent Reliability
AgentOps, multi-agentic system design, and evaluation frameworks that measure what matters — beyond LLM-as-judge.
Alignment & Safety Engineering
Responsible AI frameworks, RLHF/DPO/GRPO-based alignment training, and guardrails & security controls.
LATEST WRITING
all posts →PROJECTS
all projects →RLM-Codelens
Architecture intelligence for large codebases. Combines AST parsing, graph analysis, and Recursive Language Models to detect anti-patterns, cycles, and layering violations. Multi-language support, tested on repos up to 3.4M LOC.
Operationalizing GenAI
Code and materials from my ODSC APAC 2024 keynote. Practical implementations of knowledge distillation, pruning, quantization, and model parallelization for production LLM deployment.
Telecom Domain Fine-Tuning
Synthetic conversation dataset generator for training telecom customer service AI. Logic-based plan suggestions, multi-turn dialogues, and built-in dataset validation with quality metrics.