A conference chatbot that answers its own questions. An AI concierge for SIM Conference 2026, trained on the official program and answering questions in seconds, using real sources.
The problem
A conference is a pile of information that changes up to the day.
SIM stands for Startups & Investment Matching, the conference Startup Portugal runs to get founders and capital into the same room. The 2026 edition ran for two days, May 14–15, in Porto, and the promise was right there in the name: where founders meet capital.
Every attendee needs a slightly different slice of the program: agenda, speakers, venue logistics, ticket tiers, how the investor 1:1s work, what to wear. And they tend to need it right now, usually on a phone, the night before.
The usual places you find answers don't hold up very well:
- A static FAQ page goes stale the moment the agenda shifts, and it never quite covers the question someone actually has.
- A long program PDF holds everything and surfaces nothing. Nobody reads 40 pages to find a room number.
- An inbox or help desk just turns into a queue, where the same twenty questions get answered by hand by a team that should be focused on running the event.
What an event actually needs is for the program itself to be answerable, so anyone can ask it anything in their own words and get a straight answer with a source attached.
How we approached it
We pointed Ardaven at the program, and almost none of it was code.
We didn't build a one-off app for a single conference. We pointed Ardaven, LayerX's chatbot platform, at the SIM program and stood the assistant up on top of it. Most of the work was curating what it knows and how it talks, not engineering a chatbot from scratch.
We made the program the source of truth
Everything the assistant references traces to the official 2026 material: the agenda, the speaker list, venue and logistics notes, ticketing, the investor-matching format. We loaded those documents into the bot's knowledge base and let the platform handle the rest. It reads them, breaks them into searchable passages, and indexes them so a question lands on the right paragraph. When the program changes, you update the document, not the code.
We made it answer with receipts
A conference assistant that guesses is worse than none at all. A wrong room number sends someone to the wrong building. So Ask SIM is grounded. It finds the relevant passages in the program first, answers from those passages, and cites where the answer came from. The line on the site, "answered, instantly, with sources," describes how the system actually works.
We made it look and feel like SIM
The greeting, the tone, the four starter questions, the accent colour, the "trained on the official 2026 program" line: all of it is configuration on the bot, with no custom code. The assistant introduces itself the way the event would, and it nudges people toward the questions they actually have.
The approach
Ground it in the real program · make it answer with sources · dress it as SIM.
What we built
One assistant, the conference's knowledge behind it.
Ask SIM is one assistant with the conference's knowledge behind it, reachable wherever an attendee already is. The screen stays deliberately plain: a greeting, four things to try, and a box that says ask about speakers, agenda, venue, tickets… Ardaven does the knowledge search underneath.
Trained on the official program
Everything the assistant knows comes from the SIM 2026 material itself. Organisers upload what they already have, including the program, agenda, venue and logistics notes, speaker bios, ticketing and FAQ documents, even web pages, and Ardaven turns each one into something the assistant can search. The formats are the ones an event already produces: PDF, text, Markdown, web links. There's no special data prep, and several documents can sit under the same bot and get searched together.
This is knowledge-based search, rather than fine-tuning.
Every answer carries its source
When someone asks a question, Ask SIM doesn't reach into a generic model's memory. It searches the program, pulls the passages that relate to the question, answers from those, and points back at where the answer came from. That's the difference between a confident-sounding guess and an answer someone can act on. Ask "where exactly is the venue?" and you get the venue from the logistics document, not a plausible-sounding hallucination.
Ask a follow-up like "and how do I get there from the airport?" and it understands that "there" means the venue you just asked about. It rewrites your question against what's already been said before it searches, so context carries the way it would in a real human interaction.
Starter questions that break the ice
A blank chat box is intimidating. People don't know what it's allowed to answer. So Ask SIM opens with four starter questions, drawn from what attendees actually wonder about. One tap and they're in, and the four are configurable per event.
It looks and sounds like the event
Everything an attendee sees is branded to SIM, with no bespoke front-end work. The header ("Ask SIM · trained on the official 2026 program"), the greeting ("Hi, I'm trained on the SIM 2026 program. Ask me anything."), the tone of voice, the accent colour, the placeholder text. All of it is set on the bot. The assistant feels like part of the conference site, because to a visitor, it is.
Live wherever attendees are
The same assistant ships three ways, and SIM can use any or all of them:
- A standalone page at sim.layerx.xyz, which you can share as a single link in an email, a ticket confirmation, or a QR code on a badge.
- An embeddable widget, a floating "ask" button that drops onto any website with a short snippet, so the assistant rides along on the event's own pages.
- An API, for teams that want the assistant inside their own app or attendee portal.
One knowledge base, one configuration, three front doors. Set it up once, then meet attendees wherever they already are.
How it works
Turning a program into something you can just ask.
You don't need to see the plumbing to trust the answers, but here's what happens underneath. It's the same machinery for any event, document set, or knowledge base.
Making a program answerable
When a document goes into the knowledge base, Ardaven:
- Reads it. Clean text out of PDFs (including scans and tables), text files, Markdown, or a web page.
- Breaks it into passages. Overlapping chunks, precise but still meaningful.
- Indexes it for meaning. Each passage becomes a vector in a vector index, so search matches on what a question means rather than its exact words.
That's why "what should I wear?" can land on a dress-code line that never uses the word "wear."
Answering a question
- The question is turned into the same kind of searchable vector.
- The most relevant passages from the program get pulled back.
- Those passages go to the model as ground truth, with an instruction to answer from them and cite the source.
- The answer streams back word by word, so it feels instant on a phone.
A follow-up gets rewritten to stand on its own first, so the search still finds the right passage. And if the knowledge step ever fails, it keeps working instead of breaking.
Built to be safe in public
A public assistant for a named event comes with guardrails from the platform. Rate limiting absorbs spikes and abuse, an origin allowlist means only approved sites can embed it, there are optional bot-protection challenges, and an AI-disclosure notice. GDPR and EU AI Act considerations are handled at the platform level.
What it delivers
The real win: questions answered with a human in the loop
| One source of truth | Every answer traces to the official 2026 program. Update the document, not the code. |
| Answers with sources | Grounded retrieval instead of guesswork, so a room number is a real room number. |
| Branded, not bolted-on | Greeting, tone, starter questions and colour all match the event. |
| Live in days, not months | No model training and no bespoke app. You curate the knowledge and configure the bot. |
The repetitive questions about agenda, venue, tickets, 1:1s, and dress code get answered the moment they're asked, which frees the organising team to run the conference instead of the inbox.
The architecture
Every layer was chosen for a reason: grounded answers over guesswork, configuration over code, safety built in from the start.
| Layer | Choice |
|---|---|
| Platform | Ardaven — LayerX's multi-bot chatbot and knowledge-search platform |
| Knowledge | Documents (PDF, text, Markdown, web pages) chunked, embedded, and stored in a vector index for semantic retrieval |
| Answering | Retrieval-augmented generation. Relevant passages retrieved, then answered from, with sources and streaming responses |
| Surfaces | Standalone chat page · embeddable website widget · API |
| Safety | Rate limiting · origin allowlist · bot-protection challenge · AI disclosure · GDPR / EU AI Act handling |
Beyond SIM
Any body of knowledge, made answerable.
This conference program was the showcase, not the limit. The same Ardaven capability fits any document people have to dig through, whether it's a product catalogue, an employee handbook, a support knowledge base, or a policy library.
Wherever there's a pile of information and a stream of people asking the same questions, you can turn it into an assistant that answers in plain language with the source attached.
Let's build your Ask SIM.
Every company has a version of this problem. Information that exists but isn't accessible, and a team fielding the same questions on repeat.
We build the AI and knowledge systems that turn a company's own information into something anyone can just talk to, pairing real depth in AI and search with modern software practice and built to be safe, secure, and compliant for the EU.
If that sounds like your situation, let's talk.
Let's Talk