FELTHOR
Imediato Nexway · logistics and transportation

One day, 99 people, 13 AI prototypes

The hackathon that opened Imediato Nexway’s AI culture program. The goal was not to award a project: it was to have people who had never used AI leave the day building with it.

99people, on site and online
85real problems submitted by participants
13prototypes pitched to the jury
95%fewer hours in the winning team’s process (team estimate)
Context

Imediato Nexway built an AI culture program on three pillars: governance, training and tooling. The October 2, 2026 hackathon was the program’s first company-wide action.

The starting point was honest: most people had never used AI at work. When the event was announced, one department thought it was about “building a little game”.

The challenge

Three things at once:

  • A mixed crowd. On the same day, people who had never opened an AI chat and people who already automate. Beginners could not be left behind, and advanced users could not get bored.
  • Real problems, not shelf prototypes. Whatever came out of the day needed an owner, a size and a path to production.
  • Security. No software installed on company machines, nothing published outside the company, and care for what already exists: about 2,000 computers in daily use.
What Felthor did
  1. Designing the day

    Inspiration, technical leveling and hands-on, in that order. A guided script from start to finish, with a 15-minute cap on each technical talk. The coffee break had a technical job: making sure everyone had the tool open and working before the hands-on.

  2. Scope agreed with information security

    Three levels of AI use: chat, an agent that automates work, and a system. The hackathon covered the first two and stopped at the prototype. Servers, databases and publishing go through the technology team after the event.

  3. Technical team trained the day before

    One hour with about 15 people who would mentor the teams: build a skill, an artifact and a scheduled task. The rule: nothing is taught on the day unless the mentors have done it first.

  4. A one-page canvas and a jury with a written yardstick

    Each team filled in problem, solution, expected result, risks and how AI helped. The jury scored six weighted criteria built on the pillars of the company’s technology committee, so the winner reaches the committee in a language the board already reads.

  5. Opening talk

    “Management first, AI second”: before automating, know which goal the project moves. An improvement that doesn’t target the goal is real, and still doesn’t change the result.

  6. The event platform

    Built by Felthor in two days: check-in, idea submission, team draw balanced by AI skill level, a canvas filled in live by each team, jury scoring from their phones, pitch order and the final podium on the big screen. Hosted on the company’s own server.

Results
13prototypes pitched in 5 minutes each, from 13 randomly drawn teams
85 of 99participants submitted a real operational problem
100 → 5 hper month to reconcile records in an HR process: the winning project
2,900 hper month, adding up the 13 teams’ estimates

The gains are the teams’ own estimates, made in a few hours. On stage, the total was presented as a hypothetical figure: R$ 2.1 million a year and 2,900 hours a month, within a potential scenario of R$ 12.8 million, not yet validated. Validation is the next step.

Where Claude came in
  • Claude was the event’s tool. The company gave every participant access to Claude for 30 days. The top three teams received permanent AI licenses.
  • The exercises used Claude’s own features: skills, artifacts, scheduled tasks and the Microsoft 365 connector, approved by information security.
  • Mentors who search alongside. The day before, only about three people on the technical team had used Claude. The guidance was not to pretend to know everything: search inside the tool, next to the participant.
  • The event platform was built with Claude Code in two days and hosted on the company’s server.
What we learned

An idea that starts with “we lack” is already a solution

“We lack WhatsApp access” is a solution in disguise. Asking for the problem without that word improved the ideas on the spot.

A strong business case beats a weak MVP

Nobody finishes a system in two hours. What the jury wants to see is who understood where AI helps.

The prototype stops at the artifact

In a corporate event, the hackathon proves the idea. Production systems go through the committee.

A jury needs a written yardstick

A subjective criterion without one becomes the evaluator’s taste. The yardstick was the company’s 2030 ambition.

Next steps
  • The winning projects were taken to Imediato Nexway’s technology committee on October 6, 2026.
  • The company announced its AI usage policy for the following week, built around a corporate tool.
  • The plan is for the event to become a training track. A second edition was promised to those who could not get a seat.
  • Another department has already asked to use the platform for a business case.
Testimonial
Reserved for the client’s testimonial.
Added only with their words and approval.