Live Agent System
A production agent system we built from scratch for NYSA’s AI arm (ai.nysa.tr) — live today. Visitors describe a task and watch intelligence work in real time: not a mockup, a real model stack running at zero monthly cost. Proof of our AI service as a working product, not a promise.See the live system — ai.nysa.tr
YEAR
2026
TAGS
Duration
Live — continuously improving
Team
NYSA
Stack
PROBLEM
The paradox of a new AI product: you need proof to win work, and work to build proof. The classic fix is fake case studies and inflated metrics — that collapses on the first real client and burns the brand. On top of that, a technical constraint: zero budget, no payment card, instant replies for every visitor, resilient failover, and abuse protection.
APPROACH
We stopped talking and started showing. On the AI arm’s homepage, visitors describe a task and watch the agent work — steps are honestly labeled, responses come from a real model (choreography, not proof). We built multi-layer failover: if the primary model stalls, the queue falls through automatically, up to seven models. Each model sits in its own free-tier bucket so the system never depends on a single provider. Trust and continuity layers: every conversation is logged anonymously (real demand intelligence), crash alerts email the team, bot protection guards free quota, personal data stays out.
RESULT
The outcome is the system running on ai.nysa.tr right now: not a mockup, a live model. · Monthly operating cost: ₺0 — built on free model tiers. · Seven-layer model failover → assistant stays up if one provider fails. · Every visitor question logged anonymously → real demand data. · Bot-protected → automated traffic cannot drain free quota. · Automatic alert on failure → the team knows before visitors do. Lesson: well-built AI speaks through a measurable, running system — not a flashy promise.