SaaS / AI
Smart CRM
A multi-tenant CRM with AI assistance built into the record, summarising history and surfacing the next action for every account.
- Next.js
- TypeScript
- PostgreSQL
- OpenAI
- +1
View Case Study
AI / SaaS
Assisted Support & Deflection System
A support platform that answers repeat questions from verified sources, drafts replies for agents, and escalates anything it cannot ground in documentation.
Overview
Support volume was growing faster than the team. Instead of a chatbot that guesses, we built a system that answers only what it can ground in the client's own documentation, cites the source on every reply, and routes everything else to a human with a draft already prepared. Agents review rather than compose, and the knowledge gaps the system finds become a prioritised content backlog.
Challenge
Around two-thirds of tickets were repeat questions already answered in the documentation, but an earlier chatbot had damaged trust by confidently inventing answers.
Ticket volume was rising faster than the team could hire.
A previous chatbot produced plausible but wrong answers and had to be withdrawn.
Documentation existed but was fragmented across three systems.
Leadership required that no customer-facing answer could be unsourced.
Solution
We made grounding non-negotiable: if an answer cannot be traced to indexed documentation, the system does not send it.
Unified three documentation sources into a single index with freshness tracking.
Constrained generation to retrieved content and attached citations to every response.
Tuned a confidence threshold against a labelled evaluation set rather than by feel.
Made escalation the default for anything uncertain, with a pre-drafted reply for the agent.
Turned unanswerable questions into a ranked backlog so the knowledge base improves continuously.
Capabilities
What AI Customer Support Platform does in day-to-day use.
Replies are generated strictly from indexed documentation, with a citation on every claim.
Below a tuned confidence threshold the system escalates rather than guessing.
Escalated conversations arrive with a suggested reply and the relevant sources attached.
Questions that could not be grounded become a ranked content backlog for the docs team.
Configurable voice, plus hard rules on topics the system must never answer autonomously.
Sampled conversations scored for accuracy and helpfulness, tracked over time.
Stack
The tools this system runs on, grouped by the role they play.
Architecture
The path a single request takes through the system, end to end.
Email or widget message enters the platform and is classified.
Relevant documentation retrieved from the unified index.
A response is generated strictly from retrieved content.
Above threshold sends automatically; below escalates with the draft.
Ungrounded questions feed the documentation backlog.
Outcome
What changed for the business after launch.
61%
Tickets resolved without an agent
Across repeat questions with a grounded, cited answer.
0
Unsourced customer-facing answers
Generation is constrained to retrieved documentation.
44%
Faster agent handling time
On escalated conversations arriving with a prepared draft.
3 → 1
Documentation sources unified
One index with freshness tracking per document.
Interface
Screens from the delivered system.
More work
Other systems built on similar foundations.
SaaS / AI
A multi-tenant CRM with AI assistance built into the record, summarising history and surfacing the next action for every account.
View Case Study
AI / Web Application
A web platform that parses resumes into structured data, matches candidates to roles with an explainable score, and generates tailored documents.
View Case Study
Next step
Tell us what your business needs to automate and we'll map out a practical build.
hello@devrox.comWe reply to every enquiry within one business day.