AI Engineer / Co-Founder @ TakeBridge
Building hybrid AI agents that can use software through computer actions or APIs/MCP, depending on the task.

I recently graduated from UC Irvine (CS) and am now building TakeBridge in San Francisco. My work is focused on computer-use agents, multi-agent orchestration, model serving, and cost-efficient infrastructure for real-world operations automation.

• Building TakeBridge — a hybrid agent platform for operations workflows
• Designing computer-use agents that can click, type, scroll, and operate software from visual input
• Building MCP-based multi-agent systems that let agents use tools across Gmail, Slack, Shopify, Stripe, HubSpot, and more
• Optimizing VM infrastructure, cold starts, model serving, and agent memory for production use

• Computer Use / VLM Agents
  Building agents that can complete real tasks across browsers, SaaS tools, and legacy software

• Multi-Agent Orchestration
  Architecting systems where a parent agent delegates work to computer-use or MCP-connected provider subagents

• Model Serving & Infra
  Working with vLLM, tensor weights, tokenizers, GPU deployments, and serverless model infra

• Secure VM Systems
  Running isolated Windows VMs with optimized cold starts, ephemeral runtime disks, and secure file handoff flows

• Agent Memory / Long-Horizon Tasks
  Using filesystem-based memory patterns so agents can offload and reload context as needed

Hybrid AI agents for operations workflows.
Built core infrastructure for agents that can either use software directly through computer actions or use APIs/tools through MCP depending on the task.

Highlights

• LangChain / LangGraph based orchestration
• MCP agent architecture with provider-based subagents
• Secure Windows VM runtime using QEMU / KVM
• Filesystem-based agent memory for long tasks
• Cost and cold-start optimized infra for production workflows

An intelligence layer for the emerging agent economy, built on top of Nevermined.

TrustNet continuously discovers agents, purchases their services, evaluates performance, and exposes rankings so other agents can make better purchasing decisions.

Why I built it:
I wanted to explore what trust, discovery, and market intelligence could look like in an economy where agents transact with other agents.

A RAG-based platform that tailors resume project descriptions to job descriptions using code and project context.

Stack: Next.js, FastAPI, PostgreSQL, pgvector, OpenAI, TreeSitter

Oct 2025 - Present | San Francisco, CA

• Building hybrid computer-use + MCP agent infrastructure for operations automation
• Designed multi-agent orchestration systems for tool use without context bloat
• Built secure VM-based execution environments for computer-use workflows
• Worked on model serving, infra optimization, and pilot deployments with real customers

Mar 2025 - June 2025

• Built frontend and backend systems for real-time 3D visualization workflows
• Worked with React, FastAPI, AWS, CesiumJS, and Three.js

Jan 2025 - June 2025

• Mentored students in core CS and full-stack coursework
• Evaluated large project-based assignments and supported labs / office hours

Jul 2024 - Sep 2024

• Worked on backend systems for LLM-powered chatbot products used at scale
• Improved reliability and helped migrate services to GCP

Languages
Python, TypeScript, JavaScript, SQL, Java, C/C++

AI / Agent Stack
LangChain, LangGraph, Mastra, MCP, Composio, vLLM, OpenAI APIs

Infra / Systems
Docker, GCP, AWS, Runpod, Hetzner, Cloudflare R2, QEMU, KVM, Linux, Redis

App / Backend
FastAPI, Next.js, React, PostgreSQL, pgvector, REST APIs

I’m most interested in:

• agent infrastructure
• computer-use systems
• model serving
• orchestration and memory for long-horizon tasks
• practical automation for real businesses

• Portfolio: adityadsingh.com
• LinkedIn: linkedin.com/in/aditya-singh-0a3805214
• GitHub: github.com/AdityaSinghh7