Internship - Research Engineer - Benchmarking and optimization of tool usage by language models. (in collaboration with MICS CentraleSupelec)

About Us

Alpic is the all-in-one MCP hosting platform.

We're entering a major technological shift where AI is replacing traditional search and beginning to act on our behalf through agents. These agents won’t rely on human interfaces, they’ll interact through AI-native protocols. The current UI-based web is too slow, costly, and fragile for agents, while also blocking key actions like payments or authentication. We believe that a new web powered by MCP servers is emerging, and that every online business will soon need an MCP server to stay relevant in this agent-driven future.

Alpic is building the infrastructure to power this new internet: an all-in-one platform where building, deploying, and monitoring MCP servers at scale feels like magic.

We aim to combine the ease of Vercel, the performance of CloudFlare, and the versatility of Kong’s API gateway - all for AI-native workflows.

Team

We’re a team of repeat founders with a successful exit with our first company: Streamroot, a P2P video streaming infrastructure tech. We are also early adopters and major contributors to MCP.

The founders have known each other and worked together for 10+ years: 4 met in college in Centrale Paris, 4 worked together at Streamroot and weathered the ups and downs of startup life.

We are a passionate and fast-moving international team with high ambitions. We value team work, the right to fail fast and learn, resilience, adaptability and humility. And we'd like to have a little fun while building the next billion dollar company!

What we're looking for

Responsibilities

This is a research internship in collaboration with the MICS research lab from CentraleSupélec. You will be working part-time at Alpic and part-time at the MICS laboratory.

The subject of your internship is Benchmarking and optimization of tool usage by language models.

You will research the different parameters having an impact of the Tool Success Rate from user prompts in a black-box environment (LLM + MCP servers), and propose different MCP server architectures to maximize success rate (potentially using Reinforcement Learning of Fine-tuning)

You will Analyze parameters impacting success rates of tool execution from user intent prompts in LLM systems (black box) + MCP Servers (prompts, context window, tool descriptions, MCP server architectures, error messages, model types, etc.)

1. Propose new MCP server architectures to maximize success rate - one approach being to add an SF or RL loop on a coding agent working on the MCP server

2. Test results in product on Alpic infrastructure.

Requirements

We are looking for a creative and pragmatic engineer who loves shipping products at the forefront of technology. You are a good fit if:

• You are currently pursuing a degree in Computer Science, Engineering or a related field from a tier 1 engineering school or university.

• You have at least one relevant experience in software development.

• You can work independently and thrive in a very fast-paced environment.

• You have excellent written and verbal communication skills (English required).

You might be an excellent fit if:

• You have experience with Typescript, React and Nest.js.

• You have experience in Research

• You’re curious about AI in general and MCP in particular and have built something in that space for fun or for profit.
  

Internship duration: at least 5 months
Possible shift to a full-time contract

Come join us to build the next-gen infra platform powering the agentic web!

(From left to right, Charles, Nikolay, Frédéric, Pierre-Louis and Erica)