Memory that carries AI context across sessions.
Keshav Kandoi
IBUILDWHATDOESN’TEXIST.
ABOUT
QUIET MIND.LOUD IDEAS.
I'm Keshav, an AI and full-stack engineer based in Bengaluru. I love turning ambitious ideas into real, useful products.
I'm drawn to the messy middle of building—questioning assumptions, experimenting, finding what breaks, and refining the details until everything comes together.
My approach
Question. Experiment. Build.
Education
Information Science & EngineeringJSS Academy of Technical Education · 2027
SKILLS
- Memorae
- WorkForge
- Python
- FastAPI
- LangGraph
- React
- React Native
- PostgreSQL
- Redis
- Next.js
- TypeScript
- Supabase
- Docker
- Qdrant
- Neo4j
PROJECTS
Meet people nearby who are ready to talk in person.
MeMorae
Memory that carries AI context across sessions.
The idea
Useful context is often lost between agent sessions. A longer prompt alone does not provide a durable, reviewable memory system.
I built a shared memory service with scoped ingestion, retrieval, and review, accessible through an API, LangChain integration, and MCP tools.
Technical approach
FastAPI accepts scoped requests and persists records in PostgreSQL. Background workers handle extraction and indexing. Qdrant, OpenSearch, and Neo4j are rebuildable retrieval projections; PostgreSQL remains the source of truth.
Key details
- Project-scoped memory with provenance and review states.
- Hybrid retrieval using vector, lexical, and graph indexes.
- MCP tools and documented LangChain/LangGraph integration.
Built with
Python · FastAPI · LangChain · LangGraph integration · MCP · PostgreSQL · Qdrant · OpenSearch · Neo4j · Redis · Next.js
WorkForge
Workflows that turn project activity into action.
The idea
Project events, scheduled work, and application integrations create repetitive coordination tasks across disconnected tools.
I built a workspace that connects GitHub, email, calendar, and Discord to workflows and scheduled actions, with a distinct human approval path for task emails.
Technical approach
A React/Vite client calls FastAPI. Supabase stores organization data and encrypted OAuth tokens. Webhooks trigger matching workflows, while scheduled jobs run inside the backend process. Gemini supports summaries and suggestions.
Key details
- Verified GitHub webhooks with duplicate-event handling.
- Scheduled actions and organization-scoped integrations.
- AI-assisted summaries and task suggestions.
Built with
React · Vite · Python · FastAPI · Supabase · Gemini · OAuth
LetsTalk
Meet people nearby who are ready to talk in person.
The idea
Being in the same place does not tell you who is open to a conversation.
LetsTalk lets people discover nearby places, signal readiness, and connect in person.
Technical approach
A React interface uses TanStack Start server functions for place discovery, profiles, and readiness. PostgreSQL stores application records through Drizzle; Better Auth manages authentication.
Key details
- Place discovery and ready status.
- QR-based connection flows.
- Profiles and authenticated sessions.
Built with
React · TypeScript · TanStack Start · PostgreSQL · Drizzle ORM · Better Auth
NexaChat
Local AI conversations and document search on your machine.
The idea
Cloud-only assistants are a poor fit when a conversation or document needs to remain on a local machine.
I built a desktop assistant with a local model path through Ollama and a document retrieval pipeline. Local mode runs inference on the device; optional cloud integrations are separate.
Technical approach
Electron hosts the React client and coordinates the local backend. FastAPI handles chat and document requests; Ollama runs local models, while the retrieval pipeline indexes document chunks in ChromaDB.
Key details
- Local model inference through Ollama.
- Document ingestion and retrieval with ChromaDB.
- An Electron shell around a React interface and Python backend.
Built with
Electron · React · TypeScript · Python · FastAPI · Ollama · LangGraph · ChromaDB
My approach
CURIOUS.BYDEFAULT.
I learn by building things I wish existed.
CONTACT
LET’S MAKE
SOMETHING HAPPEN.
Have an idea or an opportunity? Leave me a note.