Babu, an elderly-companion assistant

A Hindi-speaking companion for older adults that remembers their life and keeps them connected.

Starting · join earlyAI Engineering

7 of 12 tasks done

1 · Purpose

Many older adults spend long hours alone, and generic voice assistants forget everything between conversations. Babu is a companion that knows the person over time: their life story, the people they love, their routines and what is happening in their family's lives. At its core is a personal world model, a structured memory of people, events, facts, preferences and time, rather than a plain chat log. The work is requirements-first: the requirements, memory design and architecture came before any technology choice, and the first build is a prototype on a laptop before an Android tablet app.

For: Families caring for older relatives, elder-care organisations, and engineers interested in memory for AI companions.

2 · Business benefits

Company that remembersConversations build on what was said days or years ago, so they feel personal.
Speaks their languageHindi and English with natural mixing, so the person never has to adapt to the machine.
Family stays closeBabu knows what is happening in the family's lives and brings it into conversation.
The person stays in controlThe older adult decides what is remembered, shared or forgotten.

3 · Status and how to join

Starting · join early. Early days: the shape is still being decided, so joiners now have the most say. Lead: Basant Choudhary.

Good first tasks this week:

  • Write a Hindi or Hinglish exchange for one everyday situation, such as a festival morning.
  • Review a sample transcript and mark anything that sounds like a machine.
  • Add a test case for who-does-he-mean, such as two people with the same name.

The code is not public yet; joiners get access.

4 · Requirements

What it must do

How well it must do it

5 · Architecture, very high level

Speech and conversation i…Hears the person and speaksreplies, starting as a typedUnderstanding stepWorks out who the person meansand what kind of request it is.Memory retrievalSearches the world model fourways and combines the results.Permission gateRemoves anything the person hasnot agreed to share in thisPrompt assemblerBuilds the model's instructionsfrom the persona, the personReply generatorMakes the single model callthat writes the reply.Memory keeperExtracts new facts from theconversation and stores themPersonal world model stor…Holds people, events, facts,memories, preferences and timewhat the persowho and what tranked memorieonly what may be uassembled instthe exchangenew facts with
Components and what each is responsible for, with no tool names: the tools come in the next section.

Also: Personal world model store → Memory retrieval: facts, links and memories · Reply generator → Speech and conversation interface: the reply

6 · Tech stack, per platform

ComponentOpen stackAzureAWSAndroid
World model storePostgreSQL with pgvector, relationships as rows and recursive queries ✓———
Meaning searchMultilingual embedding model (BGE-M3) ✓———
Schema changesAlembic migrations ✓———
Testspytest with guarantee, retrieval and tone suites ✓———
Reply modelProposal: a hosted model with no-training terms, or self-hosted open weights———
VoiceProposal: speech-to-text and text-to-speech around the same chat loopProposal: Azure AI Speech for HindiProposal: Amazon Transcribe and Polly—
Store—Proposal: Azure Database for PostgreSQL with pgvectorProposal: Amazon RDS for PostgreSQL with pgvector—
Model—Proposal: Azure-hosted model with no-training termsProposal: Amazon Bedrock—
App———Proposal: a tablet app with large text and voice first, after the laptop prototype
On-device voice———Proposal: Android speech services, only if they pass a Hindi quality test

Bold with ✓ = built and running today. Everything else is the planned or possible binding: the components stay the same, only the tools change.

7 · What needs to be done

Requirements and design 3 of 3 done

TaskStatusSizeSkill
Write and review the requirements catalogue, with a test for every must-havedoneLrequirements
Design the memory ontology and canonical schemasdoneLdata modelling
Choose the store and write the architecture and database designdoneMarchitecture

Memory core prototype 4 of 4 done

TaskStatusSizeSkill
Schema, write path and true deletion, with testsdoneMPostgreSQL, Python
Seed a realistic fictional life of about 200 recordsdoneMwriting, data
Four-way retrieval, permission gate and conflict handlingdoneLsearch, Python
Understanding step: who is meant and what is askeddoneMNLP

Conversation prototype 0 of 3 done

TaskStatusSizeSkill
One-hour tone probe on real model repliestodoSconversation design
Prompt assembly, reply generation and memory keepingtodoMLLM engineering
Tone harness that fails the build on banned phrasingtodoStesting

Voice 0 of 1 done

TaskStatusSizeSkill
Add speech in and out around the chat loop and test Hindi voice qualitytodoLspeech

Android tablet MVP 0 of 1 done

TaskStatusSizeSkill
Build the tablet app on the proven prototypetodoLAndroid

Size: S = an evening or two, M = about a week part-time, L = a few weeks.

8 · Roles

RoleHow manyHours / weekWhat they do
Conversation designer12–3 hours/weekWrites and judges Hindi exchanges and the tone rules.
Backend engineer14–6 hours/weekBuilds the reply pipeline and memory keeping.
Speech engineer12–4 hours/weekTests and integrates Hindi speech in and out.
Android developer14 hours/weekBuilds the tablet app once the prototype is proven.

9 · What you'll learn

Level-up questions

10 · Links