An AI chatbot built on the department's own records turns hours of manual document searching into answers delivered in seconds - with every answer traceable to its source.
The client is a municipal government agency in Vietnam. Like most government offices, its daily operations depend on a large volume of documentation - internal reports, procedures, and official correspondence that staff need to reference constantly to do their jobs and respond to inquiries.
The scope of the project covered all incoming and outgoing documents the department processed within a one-year period - more than 5,500 PDF files in total, ranging from typed reports to scanned images of paper records.
Employees at the agency relied on manual methods to find information buried across this document repository - reading through files themselves or asking colleagues who might remember where something was recorded. There was no structured way to search across the full document set, summarize what a document contained, or check whether a similar issue had already been handled before.
This slowed down the department's core administrative work. Every routine inquiry that required digging through paperwork pulled staff away from higher-value tasks, and without a reliable way to verify an answer against its source document, employees had no consistent way to confirm they had found the right information. Institutional knowledge of similar, previously handled cases stayed locked inside individual employees' memory rather than the department's own records, where it could easily be lost or simply overlooked.
The department set three specific requirements for a solution: an AI agent that could summarize document content, a chatbot that could answer questions about those documents, and the ability to reference similar documents or issues that had already been processed.
|
PROBLEM More than 5,500 PDF documents - reports, procedures, and correspondence - with no fast way to search, summarize, or cross-reference them. |
CONSEQUENCE Staff lost hours to manual searching and repeated questions to colleagues, with no reliable way to verify answers or draw on how similar cases had been handled before. |
Unitech DACH built an AI chatbot purpose-designed for document administration and retrieval, built around the department's three stated objectives: accurate answers, fast search, and clear source references.
Staff can now ask questions in natural language and get answers pulled directly from the agency's own document repository in seconds, rather than the hours previously spent searching manually. Every answer is grounded in, and linked back to, its source document, so staff can verify the response rather than take it on faith. Because the chatbot can also reference documents already on file, it surfaces similar cases or issues that were previously processed, giving officers a form of institutional memory instead of starting each inquiry from scratch. The system was designed to be embedded into the agency's existing internal systems rather than requiring staff to learn a separate, disconnected tool.
| OLD PROCESS | NEW PROCESS WITH AI CHATBOT | |
|---|---|---|
| Finding information | Search files manually or ask a colleague | Ask the chatbot and get an answer in seconds |
| Verifying an answer | No consistent way to confirm accuracy | Every answer links back to its source document |
| Handling a familiar issue | Start each inquiry from scratch | Chatbot surfaces similar, previously processed cases |
| Document formats | Manual review of PDFs and scanned paper records | OCR brings scanned documents into the same searchable system |
The system was delivered as a three-stage pipeline behind a chatbot interface:
The chatbot was deployed as a web/app interface designed to embed into the agency's existing internal systems, and built on an open, self-hostable technology stack - LangChain and FastAPI for the application layer, OpenSearch and PostgreSQL for search and storage, and a locally run open-source language model (from a choice of models such as LLaMA, Mistral, or Qwen). This keeps the agency's documents and staff queries inside infrastructure the department controls, rather than routed through a third-party API.
|
30-50% Less time on routine inquiries |
15-20 hrs Saved per week on common questions |
24/7 Support availability (was 8am-5pm) |
30-40% Faster onboarding for new staff |
Since deploying the chatbot, the agency has cut the time staff spend on routine document inquiries by 30-50%, saving an estimated 15-20 hours per week that would otherwise go to manual searching. Support that was previously limited to business hours (8am-5pm) is now available around the clock. New staff reach full productivity 30-40% faster, because they can query the department's documented procedures directly instead of relying on colleagues to walk them through it.
Reported results also include a 20-30% increase in employee engagement scores and a 25% increase in new-hire satisfaction, alongside an estimated 30-50% reduction in operational costs - freeing the department to focus more of its time on higher-value, strategic work rather than routine lookups.
| METRIC | BEFORE | AFTER |
|---|---|---|
| Time on routine inquiries | Baseline | 30-50% lower |
| Staff hours on common questions | - | 15-20 hours/week saved |
| Support availability | Business hours (8am-5pm) | 24/7 |
| New-staff onboarding time | Baseline | 30-40% faster |
| Operational costs | Baseline | Estimated 30-50% lower |
|
DEPLOYMENT OPTION Also available as a ready-to-use virtual appliance — a pre-configured virtual disk image with every required component already installed and set up — for organizations that want a fast, self-contained rollout with minimal in-house setup work. A key differentiator for customers who value deployment simplicity. |
INDICATIVE INVESTMENT Typically a fraction of the cost and time required to build and maintain a comparable system in-house. |
LangChain, FastAPI, OpenSearch, PostgreSQL, OCR pipeline for scanned documents, locally run open-source LLM (e.g., LLaMA, Mistral, Qwen), web/app chatbot frontend.