Deep technical write-ups from the DialNexa engineering team: voice AI architecture, real-time latency tuning, telephony infrastructure, speech pipelines, and the tradeoffs behind shipping production voice agents at scale.
For about a year, our AI voice agents made and received phone calls the way most products do at that stage: we rented them. A… Continue reading Bring Your Own Carrier: Building SIP Trunking In a Live Voice AI Product
In Part 1 we made the case for asking instead of clicking: nearly a hundred Dialnexa operations that Claude or ChatGPT can now handle for… Continue reading How to Connect Dialnexa’s MCP Server to Claude, ChatGPT, or Your Own Code
It’s 9:40 on a Tuesday morning and you want to know one thing: how did yesterday’s campaign do? So you log in to the dashboard…. Continue reading MCP Turns Your Voice AI Dashboard Into a Conversation
This post has two halves. The Experience is what our call transfer actually does, and why the obvious approach fails. The Engineering is the deep dive: the races, the… Continue reading What happens in the six seconds after your DialNexa AI agent says “let me get you a person”
Three matchers in, we learned the real lessons the hard way: cosine similarity that couldn’t tell “yeah” from “nah”, a base model that ranked beautifully… Continue reading The 30-Millisecond Judge: Training a Small Model to Decide When Our Voice Agent Is Allowed to Cheat
Two independent gap detectors, a nightly digest, cost as a behavioural signal, and an honest audit of which feedback loops actually close and which terminate in a table nobody reads.
Detached runs, Redis session locks with database fallback, SSE replay buffers, repairing unpaired tool calls after a restart, protocol-safe compaction, and timezone-correct scheduled autonomy.
Six rungs of agent agency, capability checks re-run at dispatch, mandatory coding plan approval with repository grounding, and why there is deliberately no merge primitive.
Why we shipped governed AI memory with zero embeddings: Postgres tsvector retrieval, OR-lexeme queries, role-scoped visibility enforced in SQL, revision snapshots, and the gaps we have not closed.
How Alfred reaches company data safely: a self-hosted MCP gateway, per-provider path allowlists, lossless response paging, and a Chrome extension with ECDSA P-256 device pairing.
