- Shipped Naptick's first production voice-to-voice AI agent end-to-end across the mobile app and embedded hardware (ESP32 / Raspberry Pi).
- Optimized the streaming pipeline to sub-second voice-to-voice latency.
- Built audio encoding and compression pipelines engineered to run on low-RAM embedded devices, for on-device sound processing inside tight memory budgets.
- Kept system components inside a ~300 KB embedded RAM footprint using careful memory management, streaming, and buffer-reuse to stay stable under severe constraints.
- Built the LLM orchestration layer with multi-agent routing, RAG over sleep data with embeddings and vector search, and conversational memory for relevant health context.
- Designed tool-calling pipelines that let agents trigger product capabilities and background workflows.
- Implemented observability for AI workflows with CloudWatch logging, event-driven pipelines, and Mixpanel analytics.
- Took major redesigns and backend rewrites from concept to production within days, keeping pace with fast-shifting direction.
Founding Engineer · Full-stack · backend, frontend, mobile, cloud & firmware · AI orchestration & multi-agent architecture
Aman
Malhotra
I build systems that scale to 200M+ monthly sessions, then own the full stack the rest of the way.
Scaled consumer fintech backend at CRED. Built production AI from zero at Naptick: LLM orchestration, RAG, and voice-to-voice agents. Comfortable across backend, cloud, mobile, and right down to the device.
- 200M+Monthly active sessions
- 150M+Monthly payments
Professional experience
Where the signal has traveled.
- Built and scaled backend systems powering 200M+ monthly active user sessions across core consumer experiences.
- Delivered infrastructure supporting 150M+ monthly payment transactions with strong reliability requirements.
- Led migration of CRED's mobile codebase from three repositories to a unified monorepo used by ~80 engineers, improving dev and QA velocity.
- Built custom migration pipelines consolidating ~500 GB of repos while preserving full git history and LFS assets.
- Introduced Protocol Buffers at the app interface layer, later expanded across business lines to improve API efficiency and contract stability.
- Drove hiring and knowledge sharing: ~80 technical interviews and sessions on Flutter multi-engine architecture, Protobuf, and secure handshakes.
- Initiated early integration with Shorebird to explore modern mobile deployment and runtime patching.
Core expertise
Three layers I go deep on.
Systems that hold under load
Distributed systems and cloud architectures, backend services, event-driven pipelines, and the release infrastructure to ship them, proven at consumer scale.
Agents that think and act
LLM orchestration and multi-agent routing, RAG over real user data with embeddings and vector search, conversational memory, and tool-calling pipelines that let agents trigger product capabilities.
End-to-end product ownership
Backend, cloud, and mobile integrated end-to-end, and when it matters, all the way down to firmware on the device. I take a feature from data through to something a person sees and feels.
By the numbers
The range, in figures.
0Monthly active sessions
0Monthly payments processed
0Embedded RAM footprint
<1sVoice-to-voice latency
0Repos migrated, history intact
0Technical interviews led
Technologies
The toolbox, by layer.
Backend & APIs
- Node.js
- NestJS
- Python
- Java
- REST
- gRPC
- Protobuf
Cloud & Infra
- AWS
- Docker
- CloudWatch
- Event-driven pipelines
- CI/CD
- Mixpanel
AI Systems
- LLM orchestration
- Multi-agent
- RAG
- Embeddings
- Vector search
- Tool-calling
Mobile
- Dart / Flutter
- Swift
- Multi-engine architecture
- Shorebird
Data
- MongoDB
- SQL
- SQLite
- Conversational memory
Firmware & Embedded
- ESP32
- Raspberry Pi
- Embedded C
- FreeRTOS
- Audio DSP
- On-device AI
Open to founding & senior roles