I build AI that listens, thinks and talks back.
I'm Mohammed Taqi Jigar. For 4.5 years I've built production voice and agent systems end to end: streaming speech pipelines, LLM agents behind MCP, and the multi-tenant platforms and UIs around them. The surface behind this text is a live spectrogram of the conversation running below.
Amber is the caller, violet is the agent. Ask a question below and the surface switches to your conversation.
Interview my work, not my adjectives.
A small agent built the way I build them: it retrieves from a fact base, cites its sources, refuses what it shouldn't answer, and shows its trace. Turn on voice and interrupt it mid-sentence.
Six channels. Two of them you can operate.
Each project lists the hard part. Agent Assist and AutoGTM include simplified, runnable models of the actual design.
Name the gap. Ship it by next review.
Every annual review reads the same way: I name a skill I'm missing, and a year later I'm running it in production.
Intern to engineer
Node.js training at Cybercom Creation, then Floatbot in May: co-browsing, live chat, a no-code bot builder.
Main contributor
Most commits on the three core codebases; 200+ bugs fixed to stabilize the console; reporting automation.
Module owner
Refresh-token login, the Super Admin backend, cognitive search, and 8 Docker images.
Voice AI and agents
InfluxDB analytics, the Agentic Flow editor, browser SIP calling, call translation, the voice gateway.
Scale and AI products
Clustered Agent Assist, Copilot and Scout agents, OAuth 2.1 for MCP, AutoGTM, ColdPegion.
One page per role.
Same facts, ordered for the job. Each opens as a PDF in Google Drive.