Skip to content
meSinghPublic

About

Fractional CTO and AI Architect. Agentic workflows, advanced RAG, low-latency voice, eval harnesses, LLMOps and token economics. 16+ yrs shipping, 6+ in Production AI.

Resources

Stars

1 star

Watchers

1 watching

Forks

Latest commit

 

History

133 Commits

Folders and files

Repository files navigation

Hi, I'm Mandeep

Enterprise AI fails in production for three reasons: untrusted outputs, runaway token costs and integrations that break at scale. I design the systems that solve all three, and I stay until they run on their own. Where AI stops being a feature and becomes a system: that is where I work.

16+ years shipping software, 6+ years in Production AI, currently Chief AI Architect at TWINii

I move companies off brittle LLM wrappers and onto production-grade AI systems: agentic workflows and multi-agent orchestration, advanced RAG with context engineering, low-latency voice, and under them the eval harnesses, programmatic guardrails, LLMOps and token economics that keep them standing once real users arrive. The part I care about is the part after the first result: the agent that still has to answer well with a tool down, in a language the evals never covered, under a budget. Right now that work is at TWINii, where we build real-time AI twins of real experts, and with a small number of seed to Series B teams I join part-time, embedded.

How I can help: start with a call, one engagement, on your team

Book a 20-minute call   How I can help

Lately, I've been working on

  • Real-time voice agents: latency budgets, interruption handling, memory that survives a session, guardrails that survive a user.
  • Multi-agent orchestration and MCP tool layers that hold their shape as the number of tools grows.
  • Eval harnesses and release gates, so a model or prompt change ships with evidence rather than hope.
  • Spec-driven development with AI-assisted engineering standards, so coding agents write inside boundaries the team agreed on first.
  • Security and supply-chain hardening across an organisation. That programme is why PolinRider Scanner exists.
  • Inference cost. I read the token bill the way I used to read the AWS bill.

The stack I architect: product and release, orchestration, retrieval and context, models and voice

Agentic systems, RAG architecture, real-time voice, multi-agent orchestration, MCP, eval harnesses, context engineering, guardrails, inference cost, LLMOps

Some of what I've built

  • TWINii: the real-time voice pipeline behind AI twins of experts, who answer in their own voice.
  • Devnagri AI: the voice and language AI platform across 22 Indian languages, as Solutions Architect and then Fractional CTO: multilingual AI workflows, LLMOps and a hybrid cloud that kept spend down while load went up.
  • Shiprocket: the core of its omnichannel commerce integrations. The company is now a unicorn.
  • HumanFirewall: first product live in 45 days, as Acting CTO.
  • Balluun: five years as Solutions Architect for a Silicon Valley B2B market network, through its cloud-native migration.

msingh.com  LinkedIn, in/mesingh9  X, @meSingh9

Tools I use daily
Claude Claude Code OpenAI Gemini Hugging Face MCP promptfoo Wispr Flow Notion Figma Playwright GitHub Actions Docker Vercel Cal.com

Mandeep Singh  ·  Delhi, India  ·  msingh.com  ·  the strips update daily from scripts/

About

Fractional CTO and AI Architect. Agentic workflows, advanced RAG, low-latency voice, eval harnesses, LLMOps and token economics. 16+ yrs shipping, 6+ in Production AI.

Resources

Stars

1 star

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages