1
1
Pakistan
3+ years of experience
AI Engineer passionate about solving diverse challenges across computer vision, deep learning, data analytics, and intelligent automation. I build end-to-end solutions from autonomous AI agents and predictive models to interactive dashboards and agentic systems that deliver actionable insights and measurable business impact. Proficient in Python, SQL, Power BI, and modern ML frameworks, I bring adaptability and a growth mindset to drive innovation across industries.

Raabta Live turns any website into a place where visitors can talk to an AI — not type. Paste one line of code, and a voice agent appears: it listens as you speak, thinks, and speaks back, in real time. The whole experience is built on AssemblyAI Universal-Streaming. As you talk, your words are transcribed sub-second and shaded by word-level confidence, and a live "time-to-first-word" meter shows AssemblyAI's speed right on screen. AssemblyAI's multilingual streaming auto-detects the language of each turn (English, Spanish, French, German, Italian, Portuguese), and keyterms boost recognition of names and brands. Because AssemblyAI does its own turn detection, the agent knows the instant you've finished — which keeps replies fast and interruptions natural. True barge-in is the heart of it: talk over the agent and it stops instantly, drops its queued speech, and listens to your new question with full context of the conversation so far. Calls end themselves after silence, and a clear hang-up button is always there. Behind the widget is a full product. Owners sign up, create agents (persona, greeting, voice, language, keyterms), and get a publishable embed key locked to their own domains. A dashboard shows live calls as they happen, a cost/usage meter, per-turn transcripts, A/B-tested personas, a voice picker with instant preview, and shareable call reports. It's multi-tenant from day one, with real auth and abuse caps. The stack is deliberately lean: one Node/TypeScript process streaming mic → AssemblyAI STT → LLM → text-to-speech over a single WebSocket, with no framework and no build step — so first audio comes back in about two seconds. It's fast because it puts AssemblyAI's streaming transcription at the center and gets out of the way.
30 Sep 2026