Skip to content

New GenAI & AI agent development

Case study · AI & ML

A support assistant that answers from real docs

A software company launched an assistant that answers customer questions using its own documentation.

  • SaaS & Technology
  • USA
  • 50–200 employees
Industry
SaaS & Technology
Region
USA
Company size
50–200 employees
Practice
AI & ML

The challenge

What was going wrong

The company's help articles, release notes and internal runbooks lived in separate places. Support agents searched them by hand, and many tickets were questions that already had a written answer somewhere.

A general chatbot was not an option. It would invent features, mix up product versions and could expose internal notes. The business needed answers that were correct, current and limited to approved content.

Our approach

How we worked it out

  1. 1

    Audit the knowledge

    We reviewed help articles, release notes and past tickets. We marked which content is public, which is internal and which is out of date.

  2. 2

    Design retrieval

    We chose a retrieval augmented approach. The model only answers from passages found in approved sources, and it says so when it cannot find one.

  3. 3

    Build the assistant

    We built the search index, the answering service and a chat widget. Answers include links to the source articles so customers can verify them.

  4. 4

    Test and improve

    We tested with real past questions, added guardrails against off-topic requests, and gave support a view of where answers fell short.

What we built

The solution

  • Knowledge index

    Approved docs are split, embedded and kept in sync with the content team's updates.

  • Grounded answering

    The model answers only from retrieved passages and cites each source.

  • Handoff to agents

    Unclear questions pass to a human with the chat history already attached.

  • Gap reports

    Support sees which questions had no good answer, so docs can be improved.

What changed

The results

  • Repeat questions handled

    Common how-to questions are answered without opening a ticket.

  • Answers customers can check

    Each reply links back to the article it came from, which builds trust.

  • Better documentation

    Gap reports show the content team exactly which topics need clearer articles.

Technology

What it was built with

  • Python Readable language for business logic, data and automation.
  • LangChain Connects language models to your data, tools and workflows.
  • Qdrant Vector database for fast similarity search.
  • FastAPI Python framework for fast, well-documented APIs.
  • React Component-based user interfaces for web apps.
  • AWS Amazon cloud hosting and services.

Keep reading

More case studies

All case studies
Manufacturing Europe

Defects caught at the station, not at the customer

Cameras catch surface defects before shipping

Problem
Inspectors checked parts by eye and missed small flaws.
What we did
Trained vision models on real parts and mounted cameras on the line.
What changed
Flawed parts are flagged the moment they pass the camera.
  • Python
  • PyTorch
  • OpenCV
  • ONNX
Read the story
Insurance UK

Claims reach the right handler on arrival

Claims emails read and sorted by software

Problem
Handlers sorted claim emails and attachments by hand.
What we did
Built a language model pipeline that reads and routes claims.
What changed
Each claim lands with the right team, already summarized.
  • Python
  • spaCy
  • Hugging Face
  • FastAPI
Read the story
Energy & Utilities Asia-Pacific

Crews hear about faults before customers do

Remote sensors warn crews before equipment fails

Problem
Remote equipment was checked by driving out on a fixed schedule.
What we did
Fitted sensors and built a live monitoring dashboard.
What changed
Crews are sent where readings show a real problem.
  • MQTT
  • Raspberry Pi
  • Espressif ESP32
  • TimescaleDB
Read the story
Taking new projects

Let’s talk about your SaaS & Technology project

Tell us about your project and get a free consultation and estimate within 24 hours.

  1. We reply within 4 business hours A real consultant reads your brief — no bot, no queue.
  2. 30-minute discovery call We dig into goals, constraints and budget. NDA signed on request.
  3. Proposal in 24 hours Scope, timeline, team and price — clear enough to decide on.

NDA on request You own the IP No obligation Free estimate

Your information is confidential. We’re happy to sign an NDA.

Book a free call Chat on WhatsApp