AVA RESEARCH & CONSULTING
DAVA
The New AI Data Engineer
A AI-powered data normalization platform that turns chaotic, multi-format datasets into unified, actionable intelligence automatically. Built for enterprise scale. Designed for human simplicity.
Why DAVA?
Most organizations drown in disjointed data — CSVs, SQL dumps, APIs, and logs that never align. DAVA changes that. It uses multi-agent LLM intelligence to generate custom parsers on the fly, process files securely inside containerized sandboxes, and deliver normalized datasets with 90 %+ accuracy.
The result:
instant interoperability, zero manual mapping, and full audit-ready visibility.
Core Capabilities
Automated Data Parsing
Generates custom parsers for CSV, JSON, SQL, TXT using large language models.
Intelligent Format Detection
Identifies schema & structure with 90%+ parsing accuracy through dual AI evaluation.
Enterprise Deduplication
Sandboxed environments for safe code execution and data privacy.
Real Time Monitoring
Web dashboards & Grafana integration for observability.
Secure Docker Execution
Sandboxed environments for safe code execution and data privacy.
System LimitsDAVA has no enforced file-size caps. Throughput depends on: Machine memory and CPU, Selected LLM model, Retry steps required for complex schemas.
The Engine Behind DAVA
-
AI-Generated Parsers:
-
DAVA’s large-language-model agents write, test, and execute format-specific parsers in real time — from CSVs to JSON, SQL, or proprietary logs.
-
-
Intelligent Format Detection:
-
Dual-layer validation ensures schema recognition with 90 %+ parsing accuracy.
-
-
Secure Execution:
-
All processes run in isolated Docker containers with zero data leakage.
-
-
Live Monitoring:
-
Grafana dashboards visualize every transformation.
-
-
LLM Redundancy:
-
Multi-provider architecture (OpenRouter, Anthropic, Ollama) guarantees uptime and fallback continuity.
-
We empower organizations to bridge complexity, scalability and speed into their data engineering.
Architecture Highlights
Installation & Setup
DAVA installs in minutes and operates fully on your local environment.
Requirements:Python 3.11+Install dependencies via uv sync or pip install -e .Configure a .env file with your preferred LLM provider key (OpenRouter, Anthropic, or Ollama)Docker optional; recommended for sandboxed executionAutomatic fallback to a secure local sandbox when Docker is unavailable
Use Case / Industries
Convert scattered audit logs into unified reporting tables.
Finance & Compliance:
Clean, deduplicate, and merge multi-source customer data.
Telecom & Retail:
Standardize incoming CSVs and XLSX forms across departments with zero engineering overhead.
Public Sector:
Deploy in 1 day. Scale to billions of records.
-
“DAVA replaced weeks of manual data cleaning with a single upload. It became our invisible engineer — fast, precise, and auditable.” — CTO, Confidential Beta Partner
Enterprise Features
Security Posture
DAVA is designed with a strict local-first model to protect sensitive data.
-
All processing occurs locally; no data leaves your environment unless explicitly configured
-
LLM-generated parser code executes in a secure sandbox (Docker recommended)
-
Simple API key authentication
-
All logs and invalid records remain local
-
No automatic deletion; retention policies are fully user-controlled
Data Storage
-
DAVA provides full transparency and control over where your data lives.
-
Raw files are stored in uploads/
-
Normalized outputs are stored in a SQLite database (normalized_data.db)
-
Smart Tables mode creates tailored SQLite table structures automatically
-
All exports are saved as CSV files inside output/jobs/{job_id}/
Upcoming Enhancements
-
Human-in-the-Loop Approval
-
Before final ingestion, DAVA presents a proposed schema for review, allowing teams to edit or approve mappings.
-
Automatic Mapping to Internal Schemas
-
DAVA aligns incoming fields with internal taxonomies for seamless integration into existing systems.
