AI Demo Showcase

OpenAI + LangChain agent with function calling for bookings, RAG over a vector index of listings, and guardrails against hallucinated price or availability.

Fine-tuned a SAM2 segmentation model on a custom dataset detects walls, doors, and windows. A rule-based engine then places furniture against the detected geometry, exposing the same output to both a React UI and a public API.

Multi-tenant by auction house: each house gets its own dashboard to upload lots, schedule auctions, and manage results. Buyers see one unified marketplace across every house on the platform. A single bidding engine handles all three auction formats with mode-specific timing and notification rules.

Recruiter dashboard for managing referral campaigns and reward tiers, employee portal for submitting candidates and tracking status, candidate pipeline that mirrors the recruiter's existing ATS. Multi-region front end on a shared backend so HR teams across countries get localised branding without forking the data layer.

Multi-tenant by subdomain: each operator's site lives at their own subdomain with their branding, pricing rules, and resource catalog. One backend handles billing, calendar sync, accounting (QuickBooks), payments (Stripe), CRM (HubSpot, Salesforce), and team comms (Slack).

Built inside the Zoom SDK so interpreters join the existing audio channel, not a side meeting. Real-time interpreter availability streamed over WebSockets so the on-demand button always reflects current capacity.

User uploads through a secure portal; images are encrypted at rest in object storage. Every view is served through an internal tracked URL layer — so the platform knows who viewed it, when, and from where, and can revoke access at any time. Third-party image-match APIs surface suspected reuses for the owner to review.

Submission portal for institutions, an editorial review workflow with multiple approval stages, and a public catalogue served from a dedicated search index. Each programme carries rich metadata for faceted discovery; the index sits separate from the content store so search stays fast as the catalogue grows.

Standalone admin microservice deployed alongside the existing booking platform. Operators sign in, add and edit properties with per-locale fields, push photos and pricing, and the main booking system picks up the changes via API — no code change in the booking codebase to support a new market.

Twilio receives the call; speech-to-text streams the caller's words to an LLM agent grounded in the business's RAG knowledge base; the agent responds via real-time TTS. Workflow automation handles bookings into Google Calendar, lead writes into the CRM, and email follow-up.

Documents are parsed (OCR if scanned), analysed for structure, and chunked using an adaptive strategy chosen per document. Embeddings land in a vector database; queries route through a multi-model RAG pipeline that picks the best LLM for the document type. Every answer carries a page-level citation back to the source.

Three role-based dashboards — facility staff (initiating consults), specialists (taking consults on-call), and QMS administrators (tracking outcomes). Two-factor auth on every login. Real-time video on HIPAA-compliant infrastructure. PHI is encrypted in transit and at rest with full audit trail.

Founders upload company documents (contracts, financials, plans, customer agreements); the agent reads them at runtime and conditions its responses on the company's actual state. A proprietary 73-step roadmap structures the founder journey; Stripe handles tiered subscriptions.

User types a career goal in plain English; an NLP layer parses intent, current role, target role, industry, and level, then ranks mentor profiles by relevance. Once a match is selected, an integrated booking flow handles calendar, time-zone, and payment so the user never leaves the app.

Customer submits their URL; a crawler builds a knowledge base from public site content. A rapid training pipeline produces a company-specific agent in minutes. The agent ships as a lightweight embeddable widget that loads on any site, customised per customer (colours, logo, tone).

Five distinct ML models — LSTM, Transformer, Hierarchical Clustering, Random Forest + Gradient Boosting ensemble, and Hidden Markov Model — run on the same five-year historical data and produce independent predictions. The signal layer converts each model's output to buy/sell/hold per horizon. Users see disagreement between models, not a fused black-box consensus.

Embeddable widget + merchant dashboard + content site, all on Next.js 15 with Payload CMS embedded for marketing content and Drizzle for billing tables.

Async FastAPI + SQLAlchemy 2.0 over PostgreSQL; a typed ActionType registry and a custom job executor dispatch each step to the right microservice.

Bring-your-own OpenAI/Gemini key for generation, Unsplash for imagery, and WordPress's native scheduled-post system for the queue.