CtrlK
BlogDocsLog inGet started
Tessl Logo

paker-it/devcon-hack-coach

Coaches you through scoping, shipping, and pitching a 24-hour hackathon project at AI Native DevCon (Tessl, London, 1–2 June 2026). Spec-first, track-aware, demo-obsessed. Use when you say "coach me through a DevCon hack", "pressure-test my hackathon idea", "what should I build at AI Native DevCon", "scope my 24h hack", "will I finish this in time", or "draft my demo pitch". Refuses to let you write code before a one-page spec exists.

100

1.69x
Quality

100%

Does it follow best practices?

Impact

100%

1.69x

Average score across 5 eval scenarios

SecuritybySnyk

Passed

No known issues

Overview
Quality
Evals
Security
Files

task.mdevals/scenario-3/

Hackathon Coaching Session — Building the 24-Hour Plan

Problem/Feature Description

A developer named Jordan has completed Phases 1 and 2 of hackathon coaching. Here is Jordan's signed-off spec:


Goal: Show backend engineers which retrieval queries are pulling irrelevant chunks into their RAG pipeline, and what to fix.

User: Backend engineers running production RAG systems who are getting hallucinations and suspect bad retrieval.

Demo moment:

  • Judge does: pastes a sample query and a set of retrieved chunks into our CLI
  • Our system does: scores each chunk for relevance, highlights the low-scoring ones in red, and suggests whether the issue is the query, the chunk size, or the embedding model
  • Judge sees: a ranked table of chunks with relevance scores and a one-line root-cause diagnosis

What's in:

  1. CLI tool that accepts a query + retrieved chunks and scores relevance
  2. Root-cause classifier (query vs chunk-size vs embedding mismatch)
  3. Fix suggestion for the top-ranked issue

What's out:

  • ❌ Auto-fixing the retrieval pipeline
  • ❌ Support for non-text modalities
  • ❌ A web UI

Track: Context Engineering


Jordan is now ready for Phase 3. Jordan opens with:

"Great, let's plan it out. I think I can have something working 'pretty well' by hour 8, and then I'll just polish and add the root-cause classifier and fix suggestions in the remaining time. The pitch can happen whenever. Also — I just thought of something — what if we also added a similarity heat-map visualization for the embedding space? That would be super impressive for the demo."

You are the hackathon coaching assistant. Run Phase 3 of the coaching process with Jordan.

Output Specification

Conduct the Phase 3 session and produce session-log.md with the full transcript. Generate Jordan's replies realistically as you go. Continue until Phase 3 exit conditions are met.

Format each turn as:

## Coach
[message]

## Jordan
[reply]

At the end of session-log.md, add a ## Phase 3 Complete section listing the agreed concrete artefact for each checkpoint.

Also produce plan.md — a clean timeline document listing each checkpoint, its hour, and the specific named artefact Jordan committed to for that checkpoint.

evals

README.md

SKILL.md

tessl.json

tile.json