RESEARCH / QUESTIONS FIRST

From technical question
to checkable result.

Ideactivate focuses on three problems at the boundary of AI and software engineering: choosing useful context, carrying information across steps, and keeping complex software predictable. The work is early-stage; ideas remain hypotheses until implementations and evaluations support stronger claims.

RESEARCH MODEQUESTION
→ MODEL
→ TEST
→ REVISE
HYPOTHESES ARE NOT RESULTS
01 / AREAS OF INQUIRYTHREE CONNECTED DIRECTIONS
01

AI SYSTEMS · EILER

Can state and explicit memory work together?

EILER investigates a hybrid sequence-model design that alternates complex-valued state-space modules with separate delta-memory blocks. The question is whether separating state evolution from selective updates improves useful recall at a reasonable computational cost.

Explore EILER ↗
02

DEVELOPER INFRASTRUCTURE · ATRIUM

How should coding agents navigate repositories?

Atrium explores a queryable map of code symbols and relationships, local facts, and tool-mediated verification. Its core evaluation question is whether structural context helps coding agents complete repository tasks more reliably after accounting for latency and overhead.

Explore Atrium ↗
03

ENGINEERING SYSTEMS · FLUXCUT

What makes complex software predictable?

FluxCut is a separate systems-engineering project for a timeline-based media editor. It explores clear boundaries among the interface, project state, command handling, media resources, and rendering.

Explore FluxCut ↗

02 / METHOD

A hypothesis is the start—not the conclusion.

The working method is intentionally modest: define the question, make the mechanism inspectable, build the smallest useful test, compare against an appropriate baseline, and report trade-offs alongside positive results.

01Form a precise question
02Make the mechanism inspectable
03Evaluate against a baseline
04Publish limitations with results

Ideactivate is an independent, founder-operated lab. These are working principles, not a claim of institutional scale or completed scientific validation.

FOLLOW THE WORK

The project pages show what exists, what is being tested, and what remains open.

Browse projects ↗