An AI joining call that knows when to hand the question to a person
MITS Global Consulting’s HR gave the same joining call to every new employee. We built an AI joining call that answers only when it is at least 70 percent sure, and hands every other question to a person.
- ClientMITS Global Consulting
- SystemAI joining call
- Delivered31-Jul-2026
- Stack.NET 8 · GPT-4o
A smaller result than it was built for, and the honest one
The repeated presentation is in use as recordings; the live AI call is paused.
70%
Confidence before the avatar answers300 to 350
People joining or changing a year5 to 10
Joiners in the live trial4
Ways to show a portal demo
Project snapshot
- Client
- MITS Global Consulting Pvt. Ltd.
- Domain
- HR & Workforce
- Engagement
- New build
- Build
- ASP.NET Core MVC on .NET 8
- AI
- Retrieval over MITS’s knowledge base, GPT-4o
- Call
- Azure Communication Services, D-ID avatar, ElevenLabs voice
What we built, and when it hands over
Aditi IT Software Solutions built an AI-led joining call for MITS Global Consulting, an IT services and staffing firm that places specialists with banks and other IT organizations, with about 700 people on its payroll. Every new joiner used to get the same orientation call from an HR person: the employee portals, the timesheet process, income tax declarations and company policies, followed by whatever questions the joiner had. With 300 to 350 people joining or changing each year, that call was repeated hundreds of times.
ConstraintThe joiner’s questions are the hard part. The scripted segments can be automated by anyone; an open question about leave, a client site or tax cannot be answered by an AI that guesses, because a confident wrong answer to a new employee becomes an HR problem later. MITS also wanted HR in the loop without making HR sit through every word, and the knowledge the answers came from had to stay on MITS’s own premises in production.
DecisionThe call has three parties: the new joiner, an AI presenter (a video avatar with a synthetic voice), and a junior HR person who watches rather than talks. Open questions are answered by retrieval over MITS’s own knowledge base and GPT-4o, which returns an answer together with a confidence score. At 70 percent or above, the avatar answers. Below it, the avatar says it will check with the HR team, gives no answer, and the question appears on the HR person’s screen in real time with its score.
The hand-off to a person was part of the design from the start: on a new joiner’s first day, a question left for HR costs less than a wrong answer. The threshold is a setting, not a constant in the code, so MITS can tune it.
BuildFour phases: HR uploads the joining letter and the system extracts the joiner’s profile; HR chooses the segments for this joiner and sends a unique link; the live call runs the segments in order (introduction, portal demonstrations, income tax, open questions, closing); afterwards a transcript and a summary are stored.
The hand-off: confidence-scored answers, the flag pushed to the HR dashboard over SignalR, and an integration test that opens a real connection to prove the flag arrives. Both paths were tested: an in-scope question answered at confidence 1.0, an off-topic one flagged at 0.2.
Portal demonstrations through four interchangeable back ends: self-hosted video, Loom, Storylane and a PDF narrated by the avatar.
Stack: ASP.NET Core MVC on .NET 8, the OpenAI API (GPT-4o and text embeddings), Azure Communication Services for audio and video, ElevenLabs voice, a D-ID avatar, Azure AI Search for development and Qdrant with SQL Server on MITS’s premises in production.
What it is built on
- ASP.NET Core MVC
- .NET 8
- OpenAI GPT-4o
- Azure Communication Services
- Azure AI Search
- Qdrant
- SQL Server
- SignalR
- D-ID
- ElevenLabs
What happened
Delivered and signed off on 31-Jul-2026. MITS ran the live AI call with 5 to 10 new joiners. Their verdict was that the content felt repetitive from one joiner to the next, and they asked for it to be generalized. Aditi recorded the AI presentation as videos, and MITS’s junior HR staff now play those recordings in joining calls. The live AI call is paused.
That is a smaller result than the system was built for, and it is the honest one. The part MITS kept is the part that was repeated word for word on every call; the part that was paused is the part that needed the joiner’s own questions. The confidence hand-off is ready for the day those questions come back into scope.
In the client’s words
Aditi IT Solutions has also helped us automate our joining formalities and induction process. During these calls, we explain key information such as employee portals, timesheet process and company policies. Earlier, our HR team had to conduct these calls manually for every new employee. The automated solution has significantly reduced human effort.
Named where the client has given written permission, described by sector where they have not.
