INDX Summer of Code 2026

An AI-native open-source internship

An AI-driven take on Google Summer of Code, for university and graduate students in Japan. Rolling start from August 2026: build software with a clear user story and real use cases, under review from full-time engineers, and publish it as open source.

You can start it as a side project alongside job hunting or research.

Per project. Start date and length are negotiable.

From ¥2,000 per hour, set by experience and skill.

Every deliverable ships as open source.

Remote or on-site, your call. No fixed hours.

INDX Summer of Code is an internship built around AI coding tools. In three weeks you finish one piece of software and publish it as open source. Getting the AI to produce something that runs is not the finish line — the scope runs through review by full-time engineers until the code is readable by others, runnable by others, and fixable by you later.

01User storyWho, what they want,and what success looks like02AI-driven buildDrive AI coding toolstoward a working build03Code reviewFull-time engineersreview once a week04Open sourceShip it so anyone can run it,then keep improving itReview and revise, week by week

Why join

01

We do not accept "the AI wrote it and it runs." Design, naming, error handling and tests get reviewed in a weekly session, with the reasoning spelled out.

  • A weekly review session (60 min)
  • A design review at each milestone
  • Ad-hoc questions any time

02

Not "something impressive with AI" — something with a user story and use cases defined up front. Concrete specs are why three weeks is enough to finish.

  • Start from a user story
  • Ship use case by use case
  • Definition of done up front

03

Everything ships under an OSS license. After the program it is still your repository to improve — and one URL on your resume points at real code and real commits.

  • MIT / Apache-2.0
  • Yours to keep improving
  • A resume URL that holds up

Reviews are done by our full-time engineers. The three below set the engineering direction and the bar your code is held to.

Katsuya Ito

01

Chief Executive Officer

伊藤 克哉

Education
Mathematics at the University of Tokyo, then the Graduate School of Economics at the same university.
Experience
Lead engineer at PFN (Preferred Networks), then Mitsui & Co., where he ran AI projects from zero to one.
Kensuke Takatani

02

Chief Operating Officer

髙谷 謙介

Education
Biology at the University of Tokyo, then the doctoral program in life sciences at the University of Tsukuba (withdrew).
Experience
Omron and FPT Consulting Japan — a background spanning manufacturing and IT consulting.
Aurélien Cluzeau

03

Chief Technology Officer

オレリアン・クルゾ

Education
Graduated from CentraleSupélec, a Grande École in Paris, France.
Experience
Founded the AI startup Buster.AI and served as its CTO. Moved to Japan after the exit and joined INDX in 2025. Ten years as a full-stack engineer, four as a startup CTO.

No paying for cloud time or AI tooling out of your own pocket. We provide what the work needs; you decide how you work.

Use the INDX cloud environment, GPUs and managed services included. No personal server bills.

We provide accounts for the AI coding tools we use in production. No personal usage caps to worry about.

Fully remote works. On-site is welcome too, and you can switch mid-program.

No fixed working hours. Fit it around classes, research and job hunting.

One engineer is assigned per project, reachable on Slack whenever you are stuck.

README, license, sample data, demo video — we see the release through with you.

Focus areas

Themes come from these four areas — all of them work we actually do, which means someone in-house can review your code. Bringing your own theme is welcome too.

01

Collecting information scattered inside and outside the company, structuring it, and making it retrievable — treated as one continuous problem.

Concrete themes

  • A base that spans meetings, chat and documents, and tracks decisions on a timeline
  • A retrieval layer agents call instead of grep, at a tenth of the token cost (MCP server)
The OSS it becomesindx-kb

Ingest several sources and stand up an internal knowledge base that answers with citations, in one command.

$ uvx indx-kb ingest ./sources
$ uvx indx-kb ask "why was this decided?"
PythonSQLiteMCPEmbedding

02

Raising small open-weight models to production quality, so data that cannot leave the building still gets processed.

Concrete themes

  • Distilling an 8B model onto a single task until it matches frontier-model accuracy
  • Quantisation and KV-cache work that keeps inference usable on machines with no GPU
The OSS it becomesindx-distill

Drop in task samples and get a specialised model plus an accuracy-and-cost comparison table.

$ uv tool install indx-distill
$ indx-distill train ./task --base qwen3-8b
PythonPyTorchvLLMLoRA

03

Handing judgement calls to an agent, so people spend their time framing the question and checking the conclusion.

Concrete themes

  • An agent that enumerates hypotheses, tests and refutes them, and returns conclusions with evidence
  • An evaluation base that records every analysis step so results stay reproducible
The OSS it becomesindx-analyst

Hand it a CSV and one question, and get a cited report whose every step you can re-run.

$ pip install indx-analyst
$ indx-analyst ask sales.csv "why are they churning?"
PythonpandasEvalsLLM API

04

Turning PDFs, forms, drawings and audio — none of it searchable or countable as it stands — into structured data.

Concrete themes

  • A form parser that survives broken layouts, with a confidence score on every extraction
  • Full-text and vector search that ships as one binary and runs offline
The OSS it becomesindx-corpus

Index an entire pile of unstructured files and compare how often any term appears, year by year.

$ indx-corpus ingest ./docs --ocr
$ indx-corpus trend "generative AI" --by year
RustDuckDBParquetOCR

Every project has a clear user story and concrete use cases — and a stranger handed the repository can start it with one command by following the README. Meet both and it is done.

REPOSITORYREADME.mdLICENSEcompose.yamlexamples/ONE COMMAND$ git clone …$ docker compose uplocalhost:3000A first-time user is running it in five minutes.Sample data included. No setup rabbit holes.
The user story fits in one sentenceTwo or three use cases actually workA first-time reader can run it from the READMEOne command to start, with sample data includedAn OSS license and a demo

The schedule is a guideline. Start date and daily rhythm bend to your calendar.

WEEK 0

Pick the theme, write the user story in one sentence, get your cloud environment and AI coding tools, create the repo.

WEEK 1

Drive the thinnest end-to-end use case with AI coding tools. Get consolidated feedback at the weekly review.

WEEK 2

Implement the rest. Design, error handling and tests get tightened through review — the steepest week technically.

WEEK 3

README, sample data, one-command start, demo. Add a license and publish it as open source.

AFTER

Keep improving it on your own. Send issues and PRs and we will keep reviewing.

Side project

Even with job hunting or another internship in motion, you can start this as a side project. Hours are shift-based, with nothing you have to attend at a fixed time. Weekends only or evenings only is fine.

And what you build gets published as open source. One URL on your resume lets an interviewer read the code and the commit history directly. Far faster than writing "I can use AI" — show what you finished and shipped with it.

A URL for your resumeCommit history as evidenceYours to keep improving

No professional experience required. We do assume you can read code yourself and understand roughly how AI coding tools work.

Required

  • Enrolled at a university or graduate school in Japan (any year, any major)
  • Real command of at least one scripting or compiled language — Python, C, Go, TypeScript, your choice
  • A working understanding of AI coding tools: skills, harnesses, workflows, worktrees, loops
  • Comfortable with Git and GitHub — branches, pull requests, responding to review
  • Willing to publish the result as open source

Nice to have

  • Code you wrote for a personal or research project
  • Experience building either a CLI tool or a web app
  • Domain knowledge — finance, healthcare, manufacturing, your research field
  • Any prior OSS issue or pull request

Even if few of these apply, apply anyway if you have something you want to build. We start with a 30-minute chat.

Outline

Who
Students enrolled at a university or graduate school in Japan (BSc, MSc, PhD)
Duration
Three weeks per project Rolling start from August 2026. Start date and length are negotiable.
Location
Fully remote, or our office at 7-7-7 Roppongi, Minato-ku, Tokyo Either works, and you can switch mid-program.
Hours
Shift-based No fixed working hours. You set your own schedule.
Contract
Part-time employment, fixed term
Pay
From ¥2,000 per hour Set according to experience and skill.
Deadline
Rolling Applications close once the places are filled.

Selection process

  1. 01Send the application formThe form is at the bottom of this page.
  2. 02Application reviewAn engineer replies within two business days.
  3. 03Casual chat (30 min, online)We talk through what you want to build.
  4. 04Align on theme and scheduleYou settle the theme with your mentor.
  5. 05Confirmed — environment handed overCloud environment and tool accounts are issued.
Q.
A.

Yes. Most people start as a side project, a few hours a week. Hours are shift-based and self-declared, and the start date is flexible. Three weeks is a guideline for focused work, not a deadline.

Q.
A.

Yes. We look at what you have built — any size — and what you want to build next. We will help you set up your AI coding workflow after you join.

Q.
A.

Yes. There are no fixed hours. Evenings only or weekends only is fine.

Q.
A.

No. We provide INDX cloud infrastructure and AI coding tool accounts. A browser and an editor are enough to start.

Q.
A.

It ships under an OSS license (MIT / Apache-2.0) and stays yours to improve and use afterwards. No customer data or confidential material goes into these projects.

Q.
A.

Application form, a 30-minute casual chat, a conversation to align on the theme, then confirmation. We cover compensation details in the chat.

Apply

An engineer will reply within two business days. We start with a 30-minute casual chat to align on what you want to build and how.

Languages and tools

Languages you can build in (multiple)*

Dev tools you have used (multiple)

In your own words

What you would build in three weeks, and who would use it. Naming one of our focus areas is enough.

The following are optional. Bullet points are fine.

  • One thing you have built, with its scale and your part in it
  • How you use AI coding tools — workflows, worktrees, loops
  • Preferred schedule, hours, remote or on-site
  • Why you are applying
  • Anything else — links to your work, your research or job-hunting schedule

Just want to talk first? That is welcome too.

contact@indx.jp