Applied AI studio · Since 2023
AIpollo Labs designs, builds and operates AI agents, agentic workflows and AI-powered apps for small and growing teams — engineered with evaluations, guardrails and a human in the loop.
Free 30-minute call · Honest advice on whether AI fits your workflow
Services
We focus on narrow, high-value workflows where AI can save hours every week — and we stay accountable for how they perform after launch.
Custom agents that reason over your data and act through your tools — for support, research, operations and back-office work.
Multi-step workflows that combine LLM reasoning, deterministic rules and human approval, wired into email, documents, CRM and ERP.
Web apps and internal tools with AI at the core — assistants, copilots and dashboards your team will actually use.
Workshops to map your processes, find the best automation candidates, estimate ROI and plan a safe, staged rollout.
How we work
Every engagement moves through four stages. You see working software early and decide what to scale based on evidence, not promises.
Map the workflow, its data and its failure modes. Agree on what success looks like and how we will measure it.
Build a focused working version on real examples, so you can judge quality before committing further.
Add evaluation suites, guardrails, logging and approval steps, and integrate securely with your systems.
Monitor accuracy, cost and latency in production, and keep improving the agent as your process evolves.
Example use cases
The kinds of systems we design and build, across functions that run on repetitive, document-heavy work.
Reads incoming tickets, drafts grounded replies from your knowledge base and routes edge cases to the right person.
Extracts fields from invoices and contracts, validates them against business rules and flags anomalies for review.
Builds concise briefs on prospects from public sources and your CRM, then drafts personalised outreach for a human to send.
Answers staff questions from policies, wikis and past work — with citations, so every answer can be checked.
Pulls numbers from spreadsheets and systems, reconciles them and drafts variance commentary for monthly reports.
Finds the right moment in video, image and document archives using natural-language queries across text and visuals.
Illustrative scenarios of what we build — not client case studies.
From the lab
Open-source systems from our team that show how we approach agents, retrieval and evaluation.
Agentic legal Q&A and contract analysis for Vietnamese law. A ReAct agent combines hybrid search (BM25 + dense retrieval with rank fusion and cross-encoder re-ranking) over 350,000+ legal documents with a Neo4j knowledge graph of amendments, replacements and citations. Answers always cite their sources, and the agent declines when evidence is insufficient.
Best configuration (hybrid + knowledge graph + re-rank) on a 200-question evaluation set.
View on GitHub →A multi-agent LLM assistant that makes market signals easier to understand. LangGraph agents handle real-time web search and personalised behavioural analysis, explainable-AI components interpret model predictions, and FastAPI services on PostgreSQL manage data streams, user state and API cost monitoring.
Built for decision support and research — not investment advice.
View on GitHub →Ways to work with us
Three engagement models, each scoped and priced to your workflow after a free discovery call.
A short, focused assessment of where AI can help most — and where it shouldn't be used.
A working agent on your real data, evaluated against agreed success metrics.
We take a proven pilot to production and keep it healthy as your business changes.
Principles
AI that touches real customers and real money has to be trustworthy. These are the rules we build by.
Team
A founding team of AI engineers and researchers from the University of Science, VNU-HCM, with peer-reviewed publications and award-winning work in retrieval, LLMs and computer vision.
AI engineer and researcher in multimodal retrieval, LLMs and OCR. Builds GraphRAG systems and AI agents end to end — from research prototype to production FastAPI services. Co-author of KPTER (SOICT 2025, Best Presentation Award) and Second Prize winner at the ICPC Southern Regional Contest 2024.
LinkedIn
AI engineer focused on multimodal agentic RAG, OCR, LLMs and computer vision. Built AI-platform backends at Acamar AI and is an AI Research Intern at Motorola Solutions Vietnam R&D Center. Co-author of KPTER (SOICT 2025) and FOCAL (ACM LSC '26).
LinkedIn
Researcher at Saigon AI Hub (VNG Group × VNU-HCM), working on image and video retrieval with vision-language models. Published at ACM LSC '26 and IEEE CBMI 2026 on object-centric attention; 6th place at the Lifelog Search Challenge 2026.
LinkedInFAQ
Small and growing businesses, startups and teams inside larger organisations who want practical AI that fits their existing tools — without building an in-house AI team first.
We are model-pragmatic and primarily build with Claude, choosing the model for each task based on quality, cost, latency and data requirements.
Through narrow scopes, evaluation suites, guardrails, logging and human approval for high-impact actions. We also design agents to escalate when they are unsure.
We access only what a workflow needs, never use your data to train models, and agree on storage and retention rules with you before any build starts.
Yes. We are based in Ho Chi Minh City, Vietnam, and work with clients in any time zone.
Contact
Tell us what you'd like to automate. We'll reply within two business days with honest thoughts on whether AI is the right fit.
