Tanishk Narula

I like building with LLMs more than training them. Mostly agentic workflows, structured outputs, tool calling, and seeing how far you can push foundation models before they break. I've spent a lot of time using LLMs as really good parsers, routing systems, and automation engines rather than just chatbots. I also like speech models and have worked with Kokoro, Qwen3-TTS, WaveNet-derived models, EchoMimic V3, and a bunch of open-source AI tooling. On the ML side, I enjoy experimentation. I've worked with Bayesian models, Gaussian models, linear regression, decision trees, random forests, bandit algorithms, exploration vs. exploitation strategies, epsilon-greedy policies, statistical heuristics, and even spent months on an RL approach that lost to a simpler heuristic model. I'm much more interested in figuring out what actually works than forcing a fancy model into production.