inferred from GitHub
lacunex hackathon 2026
Non-commercial entry in Anthropic's Built with Opus 4.7 hackathon (April 2026): adaptive interviews with live cross-turn reasoning and a takeaway for both sides. Kept as submitted.
Core trick
Generates each interview question from full session history in a live cross-turn loop, returning a takeaway to both parties.
Evidence
README matched "Built with Opus 4.7": > **Non-commercial hackathon entry.** Built end-to-end during Anthropic's *"Built with Opus 4.7"* hackathon (Cerebral Valley, 21–26 April 2026). About 20,000 people applied, 500 were selected to take part, and 288 projects were submitted and judged. This is one of them.
README excerpt
# Lacunex (hackathon archive · April 2026) > **Non-commercial hackathon entry.** Built end-to-end during Anthropic's *"Built with Opus 4.7"* hackathon (Cerebral Valley, 21–26 April 2026). About 20,000 people applied, 500 were selected to take part, and 288 projects were submitted and judged. This is one of them. > > Lacunex is not a product or a service, and nothing here is for sale. The hosted demo is switched off. The code, the [making-of](MAKING_OF.md) and the test fixtures stay here as submitted, under the MIT licence. --- > **Cross-turn reasoning, rendered live. Both sides leave with something.** Goal-directed, adaptive interviews at scale. A Host (subject-matter expert) sets interview objectives; the platform runs every conversation live — generating each question from the full session state, producing structured insight *during* the session rather than in an overnight report, and, at session close, handing the participant a reflective takeaway worth keeping. Built on **three of the five workflow patterns** named in Anthropic's [*Building Effective Agents*](https://www.anthropic.com/engineering/building-effective-agents) — **Orchestrator-Workers**, **Parallelization**, and **Routing** — running together on every turn (see [Architecture](#architecture)). → **[MAKING_OF.md](MAKING_OF.md)** — architecture decisions, day-by-day build log, calibrated claims, what Opus 4.7 enables that smaller models don't. --- ## What it does A Host describes what they want to learn