Multi-Persona AI Avatar System

Turning years of expert material into AI personas that answer from each expert’s own knowledge, with sources users can check.

RAG developmentAI knowledge baseAI assistant developmentRetrieval-augmented generationSource-grounded AI

Client

Confidential AI Product Client

Status

Launched

Expert knowledge turned into answers users can verify

The client had a large collection of expert material across videos, documents, websites, presentations, spreadsheets, and transcripts. The product needed to make that knowledge easy to ask questions about while keeping each answer connected to the correct expert and source material.

The platform also needed to support multiple expert personas without turning every new expert into a separate engineering project. The client needed one foundation that could be reused for knowledge intake, search, answer generation, evaluation, and ongoing management as more personas were added.

Product context

The system had to support multiple experts on one shared foundation without mixing their knowledge, while keeping the source trail behind every answer clear.

Challenge

The challenge

The source material came in many formats and was not organized in one consistent way. A useful answer might depend on information buried in a long video, PDF, slide deck, website, or transcript. The system needed to prepare all of that material, preserve useful context, keep each expert’s knowledge separate, and find the right information when a user asked a question.

What we built

We built one shared system that could prepare mixed source material, keep each expert’s knowledge separate, find the right information for each question, show where answers came from, and test quality before release.

One intake process for many source types

The system processes YouTube material, websites, PDFs, slides, documents, spreadsheets, and OCR-heavy files through the same repeatable workflow.

Separate knowledge for each expert

Each AI persona has its own workspace and source library, so material from one expert does not get mixed into another expert’s answers.

Answers connected to sources

The system combines different search methods, query rewriting, reranking, and citations to find the most useful source passages and show where an answer came from.

Quality checks before release

Question sets are run against the system and turned into reviewable reports so weak answers can be found before a new version reaches users.

Result

The result

The system now supports 10 expert personas on one shared foundation, with more than 20 GB of source material prepared and a clear process for checking answer quality before release.

The client now has a repeatable way to turn large expert archives into interactive AI products. Existing personas can be maintained on the same system, and additional expert knowledge can follow the same intake, retrieval, evaluation, and management process instead of requiring a separate product foundation.

20+ GB

expert source material prepared for the system

expert personas supported on the same foundation

Hybrid

retrieval combining dense search, lexical matching, and reranking

Versioned

evaluation reports used to check answer quality before release

Client feedback

“The project needed strong AI thinking, clean implementation, and careful handling of complex requirements. Ascent Innovate turned the idea into a structured system and suggested improvements that made the final product stronger.”

Name withheld

Product Lead, Private AI Platform

The impact

Why this mattered

The value came from what these decisions changed for the people using the product and the team responsible for running it.

Expert knowledge became easier to use

Years of expert material can be searched and questioned through AI instead of remaining spread across static files and media.

Users can check the sources behind answers

Citations connect responses back to source material, so users can see where the information came from instead of relying only on how confident the AI sounds.

More experts do not require a new foundation

The same core system can support additional expert personas, giving the client a clearer path to expand the product without rebuilding knowledge intake, retrieval, and evaluation from the beginning.

More case studies