The Kick­off: Six Minds, One Mission

On April 24, 2026, we—six app­ren­ti­ces and trai­nees from the field of com­pu­ter science—took part in the KI­cka­thon in Heil­bronn. The KI­cka­thon brings teams tog­e­ther to ta­ck­le real-world chal­lenges re­la­ted to ar­ti­fi­ci­al in­tel­li­gence over the cour­se of an in­ten­si­ve day. For us, it was a gre­at op­por­tu­ni­ty to put theo­re­ti­cal know­ledge into prac­ti­ce while working with mo­dern AI platforms.

Our Task: The AI On­boar­ding Chatbot

Our chall­enge re­vol­ved around a ques­ti­on that is high­ly re­le­vant for many com­pa­nies: How ef­fec­tively can dif­fe­rent lar­ge lan­guage mo­dels (LLMs) be used as on­boar­ding chatbots?

The idea be­hind it is simp­le but powerful—new em­ployees are of­ten faced with a flood of do­cu­men­ta­ti­on. An in­tel­li­gent chat­bot that can pro­vi­de di­rect, con­text-awa­re ans­wers could si­gni­fi­cant­ly ease this transition.

Our spe­ci­fic ap­proach: We fed se­ve­ral LLMs with in­ter­nal Con­fluence pa­ges as a know­ledge base and then de­ve­lo­ped an eva­lua­ti­on frame­work to compa­re their per­for­mance ba­sed on de­fi­ned me­trics. The en­ti­re de­ve­lo­p­ment and eva­lua­ti­on pro­cess took place in Azu­re AI Foundry—a mo­dern plat­form that pro­vi­ded us with the ne­ces­sa­ry in­fra­struc­tu­re for this comparison.

How the Day Unfolded

The day was tight­ly sche­du­led and re­qui­red quick de­cis­i­on-ma­king. Af­ter a brief in­tro­duc­tion, we jum­ped straight into the tech­ni­cal im­ple­men­ta­ti­on. Wi­thin the team, we di­vi­ded tasks ef­fi­ci­ent­ly: while part of the group pre­pared the Con­fluence con­tent and in­te­gra­ted it as a know­ledge base for the mo­dels, others fo­cu­sed on de­fi­ning eva­lua­ti­on me­trics and buil­ding the ana­ly­sis lo­gic. At the end of the day, all teams pre­sen­ted their results—an ex­cel­lent for­mat that also en­cou­ra­ged ex­ch­an­ge with other participants.

Con­clu­si­on: Lear­ned, Grown, Motivated

We didn’t make it into the top three—but that was far from the most im­portant out­co­me. What re­mains is a much deeper un­der­stan­ding of how LLMs can be eva­lua­ted and ap­pli­ed in en­ter­pri­se con­texts, along with hands-on ex­pe­ri­ence using Azu­re AI Foundry as a de­ve­lo­p­ment environment.

Equal­ly va­luable was the at­mo­sphe­re of the day: a team working clo­se­ly tog­e­ther, de­ve­lo­ping crea­ti­ve so­lu­ti­ons un­der time pres­su­re, and ul­ti­m­ate­ly de­li­ve­ring tan­gi­ble, pre­sen­ta­ble re­sults. Ex­pe­ri­en­ces like the­se lea­ve a las­ting impression—and mo­ti­va­te us to dive even deeper into the world of AI.

We’re al­re­a­dy loo­king for­ward to the next time.