Raman Shrivastava
Partner & Chief AI Officer, GTMX Ventures · Exited Founder · Paris
I build AI agent systems that work in production. Not demos, not proofs of concept; systems that handle real users and real money without someone watching them.
Based in Paris, working globally. I embed in engineering teams as a fractional Head of AI, shipping code and making architecture decisions alongside your engineers.
- now
GTMX Ventures
Partner & Chief AI Officer
- before
INFINIT
AI Lead, 2024 to 2025
CustomerGlu
Co-founder, acquired 2026
- focus
Agent engineering
Agentic engineering
Agent platforms
0
years building AI
0
startups co-founded
0
exit (acquired)
Acquired by Capillary Technologies, a large public company headquartered in Singapore
0
events monthly
01/what I do
three pillars
Three sides of the same problem.
The model is the easy part. Everything that makes AI useful in production lives on either side of it.
agent engineering
Building AI agents as products.
The voice agent that handles your customer calls, the video agent that processes KYC, the RAG system that serves your search. I design these across LangGraph, OpenAI Agents SDK, and Claude Agent SDK, choosing based on what the problem actually needs. The hard parts are never the model. They are context engineering, state management, observability, and graceful degradation; the 80% of the work that doesn't make it into demo videos.
agentic engineering
Making your team measurably faster.
Helping your engineers ship better software faster using AI-native development setups. CLAUDE.md constitutions, custom skills, spec-driven development, parallel agents across git worktrees, sandboxed execution. I've built and open-sourced these setups. Autonomous coding agents are still early, but the results are promising enough that I think this becomes standard within a year.
agent infra & platforms
Building the internal agent platform.
The registry, observability, and dashboard layer that lets multiple teams operate agents from one shared source of truth. Pydantic manifests per agent, multi-runtime catalogs (Vercel AI SDK, Claude Agent SDK, Lambda, Pulumi-managed GKE), Langfuse + OpenTelemetry wired identically across every deployment, a FastAPI dashboard for cross-firm visibility. Plus a curated capability registry, so agents on any runtime search and install skills via MCP, and a tool built for one team isn't reinvented by the next. The “agent platform engineer” role for companies that don't have one yet.
02/what I've built
2016 to present
From shipped systems, not slideware.
Twelve years across four seats: founder, operator, investor, advisor. Every number below came out of something that shipped.
GTMX Ventures
Partner & Chief AI Officer · Aug 2026 to present
Building agents for investing: AI strategy and agent platforms across the portfolio, the operator seat on the investor side of the table.
INFINIT
Lightspeed-backed neo-bank · AI Lead · 2024 to 2025
Built the entire AI platform from zero on Azure AI Foundry: voice agents handling 1,000+ monthly customer interactions, a LangGraph video agent on Tavus that cut KYC from hours to minutes, RAG-based car search on pgvector with sub-second responses.
CustomerGlu (Marax AI)
2016 to 2023 · Techstars '21 · Co-founder & Head of Data & AI · Acquired by Capillary 2026
Eight years, two acts. First: an AI-powered churn-reduction platform backed by Google's former head of search. RNN user-behavior prediction, LIME/SHAP explainability, action recommendation with Deep Q-Networks and contextual bandits (3× retention lift), and a GPT-2 fine-tune writing marketing copy in 2019, before ChatGPT existed. Then: a low-code platform for gamification campaigns at scale, with a GenAI layer that cut campaign creation from two weeks to ten minutes: 1B+ events processed monthly, 100+ monthly campaigns driving $20M+ in annual transactions across 30 clients, team scaled from 3 to 25+ engineers. Acquired by Capillary Technologies, a $0.5B public company, in 2026.
Seed fund
Investment Manager · 15+ AI startups advised · $12M+ raised
Technical advisor at a seed fund and innovation university in Bengaluru. 60% portfolio survival rate. I've seen AI products from every angle: building, funding, scaling, and fixing.
03/write / ask / contact
paris, working globally
Tell me what you're building.
Direct line
I write about the hard parts of building AI systems. You can also ask my AI anything about my work.