Selected AI Work

LLM inference · Human-in-the-loop agents · Robotics

AI-Powered Remote Robotic Presence Platform

Client work at Softvil Technologies · 2024 – present

A real-time robotics platform that lets people be present somewhere else through a robot, with AI in the loop. It runs in production for enterprise clients around the world.

Role
AI Engineer · Lead Developer
Year
2024 —
Focus
LLM inference
Stack
7 technologies
  • Python
  • FastAPI
  • GPT-4o
  • WebRTC
  • AWS
  • Redis
  • PostgreSQL

  • <100ms

    Control round-trip

  • GPT-4o

    LLM inference layer

  • Global

    Enterprise deployments

Overview

Remote operators see, move, and speak through a physical robot in real time. An LLM inference layer helps operators during sessions, and agent workflows automate parts of the interaction. Sensitive actions stay under human control: an operator approves them before they run.

I work across the AI and real-time layers. That covers the LLM inference service, the human-in-the-loop agent workflows, the low-latency WebRTC control pipeline, and the operator app.

Architecture

System overview

A high-level view. Client-specific details are left out.

  1. 01

    Operator app

    Operators drive the robot and talk through it in real time.

  2. 02

    Real-time transport

    WebRTC media and control channels tuned for sub-100ms round trips.

  3. 03

    LLM inference layer

    FastAPI services that call GPT-4o to assist operators during live sessions.

  4. 04

    Human-in-the-loop agents

    Agents propose actions, and operators review and approve them before execution.

  5. 05

    State & storage

    Redis for live session state and PostgreSQL for persistent data, running on AWS.

What I built

  • 01The LLM inference layer and its integration with live robot sessions
  • 02Human-in-the-loop agent workflows with operator approval steps
  • 03Sub-100ms WebRTC control pipelines
  • 04Real-time backend services on FastAPI, Redis, PostgreSQL, and AWS
  • 05The operator-facing app