Notes
Writing on voice agents, tokenizers and the tools around them.
- How far can Go go: building an optimized LLM tokenizerMy Medium write-up on OmniToken, a pure-Go LLM tokenizer, and how far Go can be pushed for tokenization speed.Medium
- How a voice agent knows you have finishedTurn detection decides when an agent starts talking. How voice activity detection and endpointing work, and why the wait setting is a trade-off.
- Designing an MCP server that spends few tokensLessons from building a Flipkart seller MCP server: fewer tools, summaries by default, capped lists, and why the tool schema is the first thing to shrink.
- OmniToken on the OpenAI Developer CommunityThe launch thread for OmniToken: the benchmarks, the supported encodings and the questions from other Go developers.OpenAI forum
- Streaming LLM responses over SSE in Node.jsHow server-sent events carry a model's tokens to the browser, with a small Node server and client that work with any streaming chat completions API.