07·AI / ML
Barnes Dubai LLM
A domain-tuned LLM and agentic broker assistant for Dubai luxury real estate, a QLoRA fine-tune on 15,369 instruction pairs, running locally.
ResultLive in the Dev Portal playground.
[ Ask about this build ]01Discover
Why it exists.
Data Centre surfaces the data; Barnes Dubai LLM reasons over it for the broker asking the question. Fine-tuned on real-estate jurisprudence, deployed locally via Ollama at zero per-query cost, reachable through WhatsApp or the Dev Portal's in-portal Barnes Dubai LLM assistant.
Brokerage-grade output needs DLD transaction conventions, France DVF reporting cadence, and Singapore URA mapping baked in, none of which live in a frontier model's pre-training in a useful way. The instruction corpus taught the base model what BARNES Dubai means by recent, comparable, or flag for review.
02Define
The brief.
Decide whether a domain fine-tune beats base + RAG for a terminology-heavy, mixed-language vertical, then ship the answer brokers can use on WhatsApp.
03Develop
What it took
Skills behind it.
Primary discipline plus the support stack.
Skills demonstrated
- AI & Machine LearningPrimary
QLoRA domain fine-tune plus a tool-calling agent loop (search_dubai_market_listings, search_barnes_own_listings, calculate_rental_yield, draft_followup_message) with output guardrails that strip ungrounded prices and fabricated listings before a reply reaches the broker.
- Prompt EngineeringSecondary
Prompt + tool-schema design for the agent loop, system prompts, function-calling contracts, a confidence-gated escalation prompt, and an eval harness over real broker queries.
- Backend EngineeringSecondary
FastAPI inference service with broker_id-keyed conversation memory; Ollama deployment; WhatsApp surface in front of the playground.
- Data PipelinesSecondary
15,369 instruction pairs from five jurisdictions' open data, with live Data Centre connectors and the Barnes listings database feeding tool calls.
- ResearchSupporting
Training-run analysis and dataset assembly that targeted brokerage-grade jurisprudence.
Develop · Notes
The build, in full.
How it came together and the decisions on the record - kept off this page so the case study stays a read, not a scroll.
Read the full build notes04Deliver
What shipped.
Live in the Dev Portal playground. A broker asks Compare Business Bay vs JVC for buy-to-let yield and gets a tool-resolved verdict in seconds (medians, gross yields, tenant pool, a one-paragraph recommendation), the brief that used to take an analyst an afternoon.
Also live via WhatsApp: the same agents behind a number a broker already messages. Under every answer sit three national property registries (Dubai DLD, UK Land Registry, France DVF) and the Barnes listings database, resolved through tool calls rather than recall.
By the numbers
Instruction pairs
QLoRA fine-tune corpus
National property registries
Dubai DLD, UK Land Registry, France DVF
Research origin
The paper-to-product thread this project sits inside. The full chain, paper through benchmark to shipped, is on the thread.
- Domain LLM fine-tuneSee thread
Teaching a general-purpose model the vocabulary of one market, on one GPU.
See also
AI / ML
Data Centre
BARNES Dubai's internal data tooling: an entity-centric valuation platform and AVM, absorbed into a broker tool over ~840,000 ownership records.
AI / ML
Infinity Broker
A white-label AI co-pilot for US luxury brokerages: one plain-language interface over the MLS, CRM and public records they already pay for.
Product
Barnes Dubai Dev Portal
The internal cockpit BARNES Dubai's innovation programme runs on: project tracking, analytics and a marketing-engineering toolkit in one place.