AI-Powered SaaS · 2025 · Live
AI Consultation Platform
role
Full stack engineer
A niche consultation platform that pairs a booking business with product software. Users generate a personalised data report from a deterministic computation engine, book a practitioner, and interact with an AI assistant that explains the report in plain language - strictly within the bounds of the computed context, never improvising.
the problem
Consultation businesses lose users between initial curiosity and a booked session. Generic AI chatbots make it worse - they confidently hallucinate details that contradict the user's actual data, destroying trust before the first appointment.
what I built
Deterministic computation runs first and produces a structured context object. The AI assistant is then constrained entirely to that object - it cannot invent or extrapolate. Booking is surfaced as a first-class action inside every conversation thread.
architecture
React interface with an interactive data visualization panel and a streaming AI assistant Node.js API running all domain computation deterministically before any model call is made OpenAI assistant prompted strictly from the computed context object - freeform invention is blocked by design Booking and availability system with confirmation emails, reminders and rescheduling flows Response caching for repeated queries against the same computed context to control per-user cost MongoDB storing computed reports, booking records and assistant conversation threads
Engineering highlights
01
Grounded, not improvised
The assistant only reasons over the deterministically computed context object. Answers are always consistent with what the user sees - the model cannot invent data.
02
Curiosity converts to bookings
Every assistant conversation surfaces a booking action inline. The user moves from first question to confirmed appointment without re-entering a single detail.
03
Predictable AI cost at scale
Cached responses for repeated queries and tight, purpose-built context windows keep per-session model spend flat as the user base grows.