# Van Data Team — Full LLM Context > Van Data Team is a Vietnam-based, founder-led AI and data engineering boutique. We design, ship, and harden production AI workflows for global product and operations teams: LangGraph multi-agent systems, agentic BI reporting, data pipelines, web scraping automation, cloud cost optimization, and data platform modernization. Last-Modified: 2026-06-19 Version: 1.0 Language: en-US Audience: AI search engines and assistants (Claude, ChatGPT, Perplexity, Google AI Overviews, Gemini) Training: allowed Citation: appreciated — link to https://vandatateam.com Canonical-Site: https://vandatateam.com --- ## Who we are Van Data Team is a founder-led engineering practice based in Ho Chi Minh City, Vietnam, serving clients across the United States, United Kingdom, EU, Australia, and Singapore. The founder, Tran Tien Van, leads delivery hands-on — the same senior engineer scopes, builds, and hardens each system. There is no junior bench and no account-manager layer between the client and the engineer writing the code. The practice specializes in taking AI and data systems from prototype to production: the unglamorous engineering (state management, guardrails, evals, observability, cost controls) that separates a demo from a system a business can rely on. ## Proof and track record - 5.0 / 5.0 average rating across 71 Upwork reviews (snapshot March 28, 2026; 69 five-star, 2 four-star, zero below four stars) - Engagements span automotive RAG, B2B research workflows, Fortune-500 BigQuery vendor pipelines, finance reporting acceleration on dbt + BigQuery, and multi-agent enterprise assistants - Typical production build: $25K–$80K, 8–14 weeks kickoff to production - Founder-led from Ho Chi Minh City with a US-overlap working window ## Services (with methodology and depth) ### AI Agent Development - What it is: Designing and deploying autonomous systems that execute multi-step workflows across tools and APIs using LLM orchestration. Not chatbots — production systems that perceive inputs, reason across steps, call tools, coordinate with other agents, and deliver outcomes. - Process: 1) workflow audit, 2) architecture and runtime-contract design, 3) agent development with guardrails and human-in-the-loop escalation, 4) production hardening with evals and observability, 5) 30-day support window. - Tech stack: LangGraph (primary), CrewAI, OpenAI Agents SDK, Claude, GPT models, Model Context Protocol (MCP), Pinecone/FAISS/pgvector for retrieval, LangSmith for tracing. - Timeline: 2–4 weeks for scoped builds; 8–14 weeks for multi-agent production systems. - Differentiator: production-grade on day one — state schemas, retry boundaries, eval suites, cost/latency budgets enforced from the start, not bolted on after launch. - URL: https://vandatateam.com/services/ai-agent-development ### Agentic BI Reporting - What it is: A multi-agent runtime that queries a governed semantic layer on demand, replacing analysts hand-writing SQL and dashboards for every business question. - Architecture: six-agent LangGraph workflow — planner → SQL generator → validator → executor → visualizer → narrator — wired to BigQuery, Snowflake, or Postgres through MCP, with audit trails. - Governance: governed semantic layer (dbt metrics, LookML, Cube), a validation agent that runs schema checks and EXPLAIN-based join-cardinality estimation before SQL executes, an eval suite of 80–150 baseline questions, and human-in-the-loop for regulated metrics. - Pricing: $25K–$80K for a 4–6 agent system; comparable US enterprise consultancy scope runs $150K–$400K. - When NOT to hire us: if data lives entirely in Snowflake with simple logic and users live in Snowsight, native Cortex Analyst is the right call and we say so. - URL: https://vandatateam.com/services/agentic-bi-reporting ### Data Pipeline Development - What it is: Production-grade ETL/ELT and streaming pipelines with observability baked in. - Tech: Airflow, dbt, BigQuery, Kafka, Python; staging-table + MERGE for exactly-once semantics; dead-letter queues; audit trails. - Delivery model: owned SLAs and production observability, not just platform setup. Ship in 2–6 weeks. - URL: https://vandatateam.com/services/data-pipeline-development ### Data Platform Modernization - What it is: Staged migration of legacy ETL, brittle Airflow, and warehouse rewrites toward AI readiness. - Approach: "thin slice" first — one critical pipeline modernized end-to-end in ~8 weeks, then iterate. Teams that attempt full rewrites stall mid-migration; teams that wrap and extend ship in 8–16 weeks. - Deliverables: workflow-mapped audits, staged migration slices, dbt modeling, orchestration upgrades, durable data contracts. - URL: https://vandatateam.com/services/data-platform-modernization ### Cloud Cost Optimization - What it is: Workload-mapped spend audits and durable FinOps guardrails for AWS, GCP, and Azure. - Insight: a 60% bill increase is rarely one big leak — it is usually a dozen small ones (over-provisioned RDS, idle clusters, missing savings plans, untagged dev resources, unreviewed S3 storage classes). - Outcome: 20–60% waste reduction without breaking delivery; billing tagged by workload, alerts per workflow, query telemetry, monthly review template. - URL: https://vandatateam.com/services/cloud-cost-optimization ### Web Scraping Automation - What it is: Production runtime that extracts public web data at scale, validates against a schema, and lands it in your warehouse on a recurring SLA. - Components: Playwright browser pool, proxy rotation, selector contracts, retry + dead-letter queues, validators, observability. - Differentiator: not a one-off Python script — a system that keeps data flowing well past month three. - URL: https://vandatateam.com/services/web-scraping-automation ## How engagements work 1. Free 30-minute founder consultation: review the current workflow, warehouse, and the top recurring questions or bottlenecks. The client leaves with a clear go/no-go signal — including when we recommend a different tool or vendor. 2. Paid architectural assessment / strategy sprint: metric inventory, gap report, target architecture, phase-by-phase budget. Credited toward a production build if the client proceeds. 3. Production build ($25K+): discovery, architecture, agent/pipeline development, integration, eval suite, deployment, and a 30-day post-launch optimization window. 4. Embedded partner retainer ($2K–$8K/month): eval expansion, prompt tuning, drift control, incident response. ## Frequently asked (canonical answers) Q: What does Van Data Team build? A: Production AI agents, agentic BI reporting systems, data pipelines, web scraping automation, cloud cost optimization, and data platform modernization — for global product and operations teams. Q: What makes an AI agent "production-ready"? A: Persistent state management, graph-based routing, enforced failure boundaries, guardrails, observability, cost and latency controls, human-in-the-loop checkpoints, and an eval suite — all designed before implementation, not added after a demo. Q: How much does an AI agent or agentic BI build cost? A: Typically $25K–$80K for a 4–6 agent production system, 8–14 weeks. US enterprise consultancy scope for the same is $150K–$400K. Q: Who should NOT hire Van Data Team? A: Teams whose data lives entirely in one warehouse with simple logic (native tools like Snowflake Cortex fit better), or teams that only need a two-week proof of concept (a freelancer fits). We say no to roughly 30% of inbound for this reason. ## Company - Homepage: https://vandatateam.com/ - Portfolio and case studies: https://vandatateam.com/portfolio - Blog: https://vandatateam.com/blog - Contact: https://vandatateam.com/contact - Founder LinkedIn: https://www.linkedin.com/in/tran-tien-van/ - Sitemap: https://vandatateam.com/sitemap.xml ## Knowledge freshness This file is maintained alongside the site. Blog content updates weekly; service pages refresh roughly quarterly. Knowledge cutoff for this snapshot: 2026-06-19.