How a Pre-IPO Investment Platform Scaled Customer Support Without Scaling Headcount
/ 3 min read
Table of Contents
About Jarsy
Jarsy is a tokenized Pre-IPO equity investment platform. Everyday investors can access shares in companies like SpaceX, Anthropic, and Stripe — starting from $10. The platform operates across Discord and Telegram, serving thousands of active users.
The Problem
Jarsy sits at the intersection of fintech, crypto, and private equity. Three domains, one compounding support nightmare.
100+ questions a day, not enough people to answer them. Tokenization mechanics, fee structures, IPO lock-up periods, proof of reserves — the small team couldn’t keep up.
Zero bandwidth beyond firefighting. Every hour on reactive support was an hour not spent on community engagement, campaigns, or user activation. They knew what to do. No capacity to do it.
Product moves faster than documentation. New features, pricing changes, policy updates. Docs written a month ago? Already outdated. Maintaining a knowledge base was a manual task that consistently fell behind.
Training a reliable human agent was unrealistic. Product complexity meant significant ramp-up time for new hires. Combined with the pace of change, the gap between “hired” and “trusted” kept widening.
The Solution
Jarsy deployed Lucius as an AI teammate to handle daily support across Discord and Telegram, with one human team member collaborating via Slack.
Lucius handles frontline support — responding to user questions across both platforms 24/7 with accurate, context-aware answers grounded in Jarsy’s product knowledge.
A human teammate collaborates via the Event System — when Lucius encounters questions it can’t resolve confidently, it creates a structured event, responds naturally to the user, and flags the issue for human review. The human confirms, corrects, or supplements — and Lucius learns from every resolution.
Continuous learning from multiple sources — Lucius doesn’t rely on a static FAQ. It learns from documentation, historical conversations, live interactions, and direct admin input — keeping its knowledge current as the product evolves.
🔥 Technical Highlight: Knowledge Conflict Detection
The biggest risk in a knowledge base isn’t duplication — duplicates just waste storage. The real danger: an old entry is wrong, and you don’t know it.
Every new piece of knowledge goes through a four-dimensional analysis:
| Dimension | What it asks | Decision |
|---|---|---|
| Content Overlap | Same topic? | High overlap → Skip |
| Information Delta | What’s new? | Has new info → Consider updating |
| Information Conflict | Do they contradict? | Determine which is correct, replace |
| Context Difference | Different scenarios? | Different contexts → Keep both |
The third dimension — information conflict — is where the real value is.
How it works:
- Vector recall + LLM precision: Embeddings for fast candidate retrieval, then LLM for semantic-level comparison
- Batched processing: Intelligent batching when candidates are many, controlling per-request context length
- Supersede chain: Old knowledge isn’t deleted — it’s archived with a record of what replaced it. Like Git history, always traceable
In practice: A user asked about the difference between SpaceX Live and Presale. Two support agents gave different answers. Both were extracted as knowledge entries. The system detected the conflict, determined the latter was more accurate, and automatically superseded the former. From then on, everyone got the right answer. No manual intervention.
Results
Always-on coverage across two platforms. Users on Discord and Telegram get responses within minutes, regardless of time zone.
Human team freed up for higher-value work. With routine queries handled, the team shifted from reactive support to proactive community engagement.
Knowledge stays current — automatically. When older knowledge contradicts newer inputs, Lucius flags it. A painful manual process became automated.
One AI + one human replaces a full support team. Jarsy avoided hiring multiple additional agents while maintaining — and improving — response quality and consistency.
In Their Words
“Lucius keeps learning and proactively alerts us when older knowledge conflicts with new information. It helps us spot what’s outdated — something we struggled to do before.” — Jarsy Team