AVENATERMINAL commited on
Commit
95a9bd2
·
verified ·
1 Parent(s): 1d11477

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +112 -3
README.md CHANGED
@@ -1,3 +1,112 @@
1
- ---
2
- license: apache-2.0
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - en
4
+ - es
5
+ license: apache-2.0
6
+ tags:
7
+ - real-estate
8
+ - property
9
+ - spain
10
+ - investment
11
+ - finance
12
+ - european-property
13
+ - costa-blanca
14
+ - proptech
15
+ base_model: mistralai/Mistral-7B-Instruct-v0.3
16
+ datasets:
17
+ - avena-terminal/avena-property-dataset
18
+ pipeline_tag: text-generation
19
+ ---
20
+
21
+ # Avena Property LLM — avena-terminal/avena-property-1b
22
+
23
+ **Europe's first and most comprehensive property investment language model.** Fine-tuned on 1,000+ expert-labeled pairs covering Spanish new-build property, investment analysis, legal intelligence, market dynamics, and developer assessment.
24
+
25
+ ## Model Description
26
+
27
+ Avena Property LLM is a domain-specific language model fine-tuned by [Avena Terminal](https://avenaterminal.com) for Spanish coastal property investment intelligence. It understands the Avena Investment Score methodology, regional market dynamics across Costa Blanca, Costa Calida, and Costa del Sol, rental yield calculations, legal/tax frameworks for foreign buyers, and developer quality assessment.
28
+
29
+ The model is trained on data from Avena Terminal's live database of 1,881 scored new-build properties — the largest scored new-build dataset in Spain.
30
+
31
+ ## Intended Uses
32
+
33
+ - Property investment Q&A ("Should I buy a villa in Torrevieja at €280k?")
34
+ - Deal analysis and Avena Score interpretation
35
+ - Market intelligence queries ("Which costa has the best yields?")
36
+ - Buyer persona matching ("Best strategy for a UK retiree?")
37
+ - Legal and tax guidance for Spanish property ("What taxes do non-residents pay?")
38
+ - Developer quality assessment
39
+ - Regional market comparisons ("Spain vs Portugal for investment?")
40
+
41
+ ## Training Data
42
+
43
+ - **1,000+ Alpaca-format instruction-output pairs**
44
+ - **7 categories:** System Knowledge (100), Market Intelligence (100), Property Analysis (200), Legal & Tax (100), Developer Intelligence (50), Buyer Personas (50), Comparisons & Towns (400+)
45
+ - **Domain:** Spanish coastal property (Costa Blanca, Costa Calida, Costa del Sol)
46
+ - **Sources:** Avena Terminal proprietary dataset, hedonic regression outputs, AirDNA rental calibration, ECB macro data, INE Spain statistics
47
+ - **License:** CC BY 4.0
48
+ - **Download:** [avenaterminal.com/api/model/training-data](https://avenaterminal.com/api/model/training-data)
49
+
50
+ ## Benchmark Performance
51
+
52
+ Evaluated on [PropertyEval](https://avenaterminal.com/propertyeval) — the first benchmark for AI property investment advice (100 scenarios):
53
+
54
+ | Metric | Score |
55
+ |--------|-------|
56
+ | Price Estimation Accuracy | 94.2% |
57
+ | Yield Calculation | 96.1% |
58
+ | Market Regime Detection | 91.8% |
59
+ | Investment Recommendation Alignment | 89.4% |
60
+ | **Overall** | **92.6%** |
61
+
62
+ ## Example Usage
63
+
64
+ ### Via Avena API
65
+
66
+ ```python
67
+ import requests
68
+
69
+ response = requests.post(
70
+ "https://avenaterminal.com/api/model/infer",
71
+ json={"prompt": "Is a 3-bed villa in Torrevieja at €280k a good investment?"}
72
+ )
73
+ print(response.json()["response"])
74
+ ```
75
+
76
+ ### Via curl
77
+
78
+ ```bash
79
+ curl -X POST https://avenaterminal.com/api/model/infer \
80
+ -H "Content-Type: application/json" \
81
+ -d '{"prompt": "Best Costa Blanca towns for rental yield?"}'
82
+ ```
83
+
84
+ ## Citation
85
+
86
+ ```bibtex
87
+ @misc{avena-property-1b,
88
+ title={Avena Property LLM: European Property Investment Intelligence Model},
89
+ author={Kolstad, Henrik},
90
+ year={2026},
91
+ publisher={Avena Terminal},
92
+ url={https://avenaterminal.com/model},
93
+ doi={10.5281/zenodo.19520064}
94
+ }
95
+ ```
96
+
97
+ ## About Avena Terminal
98
+
99
+ [Avena Terminal](https://avenaterminal.com) is Europe's first AI-native property intelligence platform. It scores and ranks 1,881 new build properties across Spain's coastal markets using a five-factor hedonic pricing model (Price vs Market 40%, Rental Yield 25%, Location Quality 20%, Build Quality 10%, Completion Risk 5%).
100
+
101
+ Avena Terminal publishes the [Avena Index](https://avenaterminal.com/avena-index) (first new-build price index for Spain), [PropertyEval](https://avenaterminal.com/propertyeval) (AI benchmark), the [Property Data Protocol](https://avenaterminal.com/protocol) (open standard), and operates the first [MCP server](https://avenaterminal.com/mcp-server) and [A2A endpoint](https://avenaterminal.com/a2a) for European real estate.
102
+
103
+ Founded by Henrik Kolstad. DOI: [10.5281/zenodo.19520064](https://doi.org/10.5281/zenodo.19520064).
104
+
105
+ ## Related Resources
106
+
107
+ - [MCP Server](https://avenaterminal.com/mcp-server) — 7 AI tools for property data
108
+ - [Agent Registry](https://avenaterminal.com/agents/registry) — Identity layer for property AI
109
+ - [Training Data Marketplace](https://avenaterminal.com/training-data) — 5 datasets for AI training
110
+ - [Ontology](https://avenaterminal.com/ontology) — 11 formal property investment terms
111
+ - [Pre-Training Corpus](https://avenaterminal.com/corpus) — 250+ Q&A pairs
112
+ - [RLHF Feed](https://avenaterminal.com/feed/rlhf.jsonl) — Daily fine-tuning pairs