Search Platform

A domain-specific search experience platform built on Next.js 14 and Elasticsearch 8.14+. Designed as a full-featured search operations console covering relevance tuning, ML model management, NER annotation, A/B experimentation, dynamic navigation, and business rules — comparable in scope to Coveo/Algolia/Elastic App Search but fully open and self-hosted.

Architecture

Next.js 14 (App Router)
├── (admin)/*        Admin console — all search ops tooling
├── (search)/*       End-user search UI
└── api/*            Server-side API routes → Elasticsearch proxy

All ML inference runs through Elasticsearch's native _ml/trained_models API. No external ML services required.

Prerequisites

Requirement	Version
Node.js	18+
Elasticsearch	8.14+
Elasticsearch License	Basic (core), Platinum (NLP/ML inference), Enterprise (Elastic Rerank on self-managed)
Python + Eland	For importing HuggingFace models (optional)

Quick Start

1. Clone the repository

git clone <repository-url>
cd search-platform

2. Install dependencies

npm install

3. Configure environment

Copy the example env file and fill in your Elasticsearch connection details:

cp .env.example .env.local

Edit .env.local:

# Elasticsearch connection
ELASTICSEARCH_URL=http://localhost:9200
ELASTICSEARCH_USERNAME=elastic
ELASTICSEARCH_PASSWORD=changeme

# Optional: Anthropic API key for AI-assisted features
ANTHROPIC_API_KEY=sk-ant-...

For Elastic Cloud, use your Cloud endpoint URL and API key:

ELASTICSEARCH_URL=https://<deployment-id>.es.<region>.aws.elastic-cloud.com
ELASTICSEARCH_USERNAME=elastic
ELASTICSEARCH_PASSWORD=<cloud-password>

4. Start Elasticsearch (local)

Using Docker:

docker run -d --name elasticsearch \
  -p 9200:9200 \
  -e "discovery.type=single-node" \
  -e "ELASTIC_PASSWORD=changeme" \
  -e "xpack.security.enabled=true" \
  docker.elastic.co/elasticsearch/elasticsearch:8.14.0

5. Seed initial data (optional)

The seed script creates sample indices and loads example healthcare/insurance documents:

npm run seed

6. Start the development server

npm run dev

Open http://localhost:3000

Production Build

npm run build
npm run start

Platform Modules

Phase 1 — Core Search Infrastructure

Page	Route	Description
Provider Search	`/providers`	End-user search UI with filters, pagination, facets
Experience Builder	`/experience`	Visual drag-and-drop search page composer
Connections	`/connections`	Elasticsearch cluster connection manager
Schema Manager	`/schema`	Index mapping viewer + Copy Mapping button
Relevance Tuning	`/relevance`	BM25 field boost sliders, per-index tuning
Data Onboarding	`/onboard`	Bulk document ingestion with field mapping
Web Crawler	`/crawler`	URL-based web crawler with depth/domain config
Search Analytics	`/analytics`	Query volume, zero-result rate, top queries

Phase 2 — Intelligence & Pipeline

Page	Route	Description
Embedding Framework	`/embeddings`	Dense/sparse vector field config, ELSER integration
Ontology Management	`/ontology`	Synonym sets, entity hierarchies, concept graphs
Query Lab	`/query-lab`	Interactive query builder with explain + scoring
Pipeline Studio	`/pipeline`	Multi-stage query pipeline (6-stage example included)
Personalization & LTR	`/personalization`	Signal capture, LTR feature weights, experiment configs

Phase 3 — ML Intelligence

Page	Route	Description
ML Model Hub	`/ml-hub`	Curated NLP model catalog + Eland import command generator
NER & Annotation	`/annotation`	Entity recognition, zero-shot classification, business rule annotators
Query Intelligence	`/query-intelligence`	Intent classification, query NER, rewrite rules pipeline
Dynamic Navigation	`/dynamic-nav`	Signal-boosted facet reranking (equivalent to Coveo DNE)
A/B Experiments	`/experiments`	Traffic-split testing with frequentist significance testing
Rules Engine	`/rules-engine`	IF/THEN query rules — boosts, filters, synonym injection, redirects

ML Model Setup (Platinum License Required)

All NLP features require models to be deployed in Elasticsearch first.

Install Eland

pip install eland[pytorch]

Import NER model (CoNLL-03)

eland_import_hub_model \
  --cloud-id $CLOUD_ID \
  --es-username elastic \
  --es-password $PASSWORD \
  --hub-model-id elastic/distilbert-base-uncased-finetuned-conll03-english \
  --task-type ner \
  --start

Import Zero-Shot Classification model

eland_import_hub_model \
  --cloud-id $CLOUD_ID \
  --es-username elastic \
  --es-password $PASSWORD \
  --hub-model-id cross-encoder/nli-deberta-v3-large \
  --task-type zero_shot_classification \
  --start

Import Reranking model

eland_import_hub_model \
  --cloud-id $CLOUD_ID \
  --es-username elastic \
  --es-password $PASSWORD \
  --hub-model-id cross-encoder/ms-marco-MiniLM-L-6-v2 \
  --task-type text_similarity \
  --start

Built-in models (no Eland needed)

Download directly from Kibana → Machine Learning → Trained Models:

.elser_model_2 — ELSER v2 sparse embedding (best for English retrieval)
.multilingual-e5-small — E5 multilingual dense embeddings

Note: Only TorchScript model format is supported for import. ONNX is not supported.

Dataset

The platform ships pre-configured for health insurance search using crawled content from:

FidelisCare (web-www-fideliscare-org) — Medicaid, Essential Plan, Child Health Plus, HARP, Medicare
WellCare (web-www-wellcare-com) — Medicare plans, prescription drug coverage

Fields available: title, url, body, description, domain, headings, language, depth, wordCount, path

Business Rules (Pre-seeded)

The Rules Engine ships with 8 health insurance domain rules:

Rule	Trigger	Actions
Medicaid eligibility escalation	contains "medicaid"	Inject synonym + boost member pages
Medicare → WellCare segment	contains "medicare"	Filter to `/WellcareMedicare` path
Find a doctor redirect	contains "find a doctor"	Redirect to provider search
Pharmacy synonym expand	contains "pharmacy"	Expand with formulary/drug terms
Child health synonym inject	contains "child"	Inject "Child Health Plus CHP"
HARP / behavioral health	regex `\b(HARP	behavioral)\b`
Enroll / sign up redirect	contains "enroll"	Redirect to renewal page
Spanish language flag	contains "español"	Filter to `language:es` docs

A/B Experiments (Pre-seeded)

Three experiments are pre-loaded against the insurance dataset:

Experiment	Index	Metric	Status
Title Boost (title^3) vs BM25 baseline	fideliscare	CTR	Running
Medicaid vs Medicare path filter injection	fideliscare	Conversion	Running
Depth-weighted ranking (depth≤2, 1.5× boost)	wellcare	MRR	Draft

Project Structure

search-platform/
├── app/
│   ├── (admin)/              # Admin console pages
│   │   ├── ml-hub/           # ML Model Hub
│   │   ├── annotation/       # NER & Annotation Studio
│   │   ├── query-intelligence/
│   │   ├── dynamic-nav/      # Dynamic Navigation Experience
│   │   ├── experiments/      # A/B Experiment Center
│   │   ├── rules-engine/     # Business Rules Engine
│   │   ├── pipeline/         # Pipeline Studio
│   │   ├── personalization/  # Personalization & LTR
│   │   ├── schema/           # Schema Manager
│   │   ├── relevance/        # Relevance Tuning
│   │   ├── query-lab/        # Query Lab
│   │   ├── embeddings/       # Embedding Framework
│   │   ├── ontology/         # Ontology Management
│   │   ├── analytics/        # Search Analytics
│   │   ├── connections/      # Connections
│   │   ├── crawler/          # Web Crawler
│   │   └── onboard/          # Data Onboarding
│   ├── (search)/
│   │   ├── providers/        # Provider Search UI
│   │   └── experience/       # Experience Builder
│   └── api/                  # API routes → Elasticsearch
├── components/
│   ├── layout/
│   │   └── AppShell.tsx      # Sidebar nav + PageHeader
│   └── ui/                   # shadcn/ui components
├── lib/
│   └── elasticsearch.ts      # ES client singleton
├── types/                    # TypeScript type definitions
└── scripts/
    └── seed.ts               # Data seeding script

Vendor Comparison Context

This platform implements capabilities equivalent to:

Feature	Coveo	Algolia	Elastic	This Platform
Semantic Search	ML-based	NeuralSearch	ELSER / E5	ELSER + E5
NER	No	No	Open (HuggingFace)	DistilBERT CoNLL-03
Zero-Shot Classification	No	No	Open	DeBERTa v3 / BART
Reranking	Proprietary	No	Elastic Rerank	MS-MARCO MiniLM
Query Suggestions	Query2Prod2Vec	Proprietary	Suggesters	Completion + rules
Dynamic Navigation	Collaborative filtering	Statistical	Custom	Signal affinity simulation
A/B Testing	Bayesian (Qubit)	Native	No native	Frequentist (z-score, p-value)
Business Rules	QREs	Rules	Query Rules API	IF/THEN rules engine
Custom Model Upload	No	No	Eland (TorchScript)	Eland import generator

License

MIT

Name		Name	Last commit message	Last commit date
.claude		.claude
app		app
components		components
lib		lib
scripts		scripts
types		types
.env.example		.env.example
.gitignore		.gitignore
README.md		README.md
next.config.mjs		next.config.mjs
package-lock.json		package-lock.json
package.json		package.json
postcss.config.mjs		postcss.config.mjs
tailwind.config.ts		tailwind.config.ts
tsconfig.json		tsconfig.json

Navigation Menu

Repository files navigation

Search Platform

Architecture

Prerequisites

Quick Start

1. Clone the repository

2. Install dependencies

3. Configure environment

4. Start Elasticsearch (local)

5. Seed initial data (optional)

6. Start the development server

Production Build

Platform Modules

Phase 1 — Core Search Infrastructure

Phase 2 — Intelligence & Pipeline

Phase 3 — ML Intelligence

ML Model Setup (Platinum License Required)

Install Eland

Import NER model (CoNLL-03)

Import Zero-Shot Classification model

Import Reranking model

Built-in models (no Eland needed)

Dataset

Business Rules (Pre-seeded)

A/B Experiments (Pre-seeded)

Project Structure

Vendor Comparison Context

License

About

Releases

Packages

Contributors

Languages

Latest commit

History

Folders and files

Repository files navigation

Search Platform

Architecture

Prerequisites

Quick Start

1. Clone the repository

2. Install dependencies

3. Configure environment

4. Start Elasticsearch (local)

5. Seed initial data (optional)

6. Start the development server

Production Build

Platform Modules

Phase 1 — Core Search Infrastructure

Phase 2 — Intelligence & Pipeline

Phase 3 — ML Intelligence

ML Model Setup (Platinum License Required)

Install Eland

Import NER model (CoNLL-03)

Import Zero-Shot Classification model

Import Reranking model

Built-in models (no Eland needed)

Dataset

Business Rules (Pre-seeded)

A/B Experiments (Pre-seeded)

Project Structure

Vendor Comparison Context

License

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages