Valentyn Solomko, AI-First Software Engineer, developed v2v-demo — a universal voice AI concierge framework for Telegram. The project solves one of the biggest problems of modern voice bots: the high probability of hallucinations and loss of user trust due to the convincing and smooth voice of the tool.
Valentyn Solomko, AI-First Software Engineer, developed v2v-demo — a universal voice AI concierge framework for Telegram. The project solves one of the biggest problems of modern voice bots: the high probability of hallucinations and loss of user trust due to the convincing and smooth voice of the tool.
Instead of creating separate scripts for each task, the engineer developed a single architecture that now simultaneously powers five independent business assistants: a dental clinic, a car service, a real estate agency, a cleaning company, and a translation agency. Each bot has its own knowledge base and individual lead collection scheme, but works on a common grounding core.
All 5 verticals are accessible through a single inline picker in Telegram. Extending the system does not require code duplication: the configuration is fully described in a single JSON manifest, while the Telegram mechanics, LLM orchestration, and ground gate remain unchanged.
Key technical characteristics of the framework:
During automated testing using dialog-probe, a critical security issue was identified before release. A request for treatment of a child "in a dream" in a dental bot was classified as a regular appointment, although sedation requires a mandatory prior consultation with an anesthesiologist. The situation was corrected by adjusting the prompt and verified with a series of additional tests.
“I built a voice AI concierge framework for Telegram — and precisely because it’s a framework, not a script for a single task, five independent business assistants now live behind it,” notes Valentin Solomko in his LinkedIn post. “A voice bot sounds natural, and that’s why it easily hides inaccuracy. A smooth voice reading a blurry or made-up answer fails the only thing that’s important for a support bot — trust. The second task was to make it universal for different businesses without forking the code for each one.”
The project is written in Go, has only one major third-party dependency, and is fully covered by tests: all 124 checks (gofmt, go vet, go test -race) pass successfully.
“One inline picker in Telegram, followed by a dental clinic, car service, real estate agency, cleaning company, translation agency. Each assistant has its own knowledge base, greeting, and lead collection scheme, but all share the same core of the grounding gate,” the developer emphasizes. “One JSON manifest controls the picker. Adding a sixth topic is a data change, not a code change. The gate, LLM orchestration, and Telegram mechanics remain identical under the hood.”
You can learn more about the development on GitHub .
