Akash Panchal

AI systems engineer

Salesforce · previously Amazon, McAfee

Akash Panchal

I design the runtimes behind production agents: state, tools, memory, voice, and the human stop before a write. Ten years of backend and product engineering. TypeScript, Python, React, FastAPI, AWS.

Experience

Work experience

Salesforce now. Amazon and McAfee before that.

AI Software Engineer, Salesforce

NowJan 2026 - PresentBengaluru, Karnataka, India

Leading the development of Agentic Content Studio - a multi-turn agent platform for Marketing Cloud.

  • Architecting the Agentic Content Studio built using React (frontend), FastAPI (python) and Mastra framework (node).
  • Designed the core runtime on Mastra's Harness primitive: per-thread state, observational memory, skills, tools catalog and its APIs, image generation and Human-in-loop.
  • Led multiple initiatives for innovation including AI calling, conversational WhatsApp and campaign building.
ReactFastAPIPythonMastra AILLMsNode.js

Senior Software Engineer - Marketing Cloud, Salesforce

Feb 2024 - Dec 2025Bengaluru, Karnataka, India

Led the modernization and integration of SMS platforms, serving billions of messages for Marketing Cloud customers.

  • Led integration of Marketing Cloud Engagement (MCE) SMS traffic with Unified SMS platform on the Falcon AWS platform — reducing aggregator onboarding time from months to weeks.
  • Spearheaded integration of a competitive SMS aggregator across LATAM and East Asian traffic, yielding ~$10M in annual customer saving. (Q3 2024 Quarter Award)
  • Architected BYO (Bring-your-own) feature for SMS aggregator, improving cost transparency and platform flexibility.
  • Proposed and implemented SMS Aggregator onboarding playbook to automate paperwork, reducing ready-to-onboard from ~16 weeks to ~4 weeks.
  • Onboarding new service for migrating existing customers from Private Data Center to Public Cloud (AWS) to provide service resiliency and reduced infra cost.
  • Providing mentorship to Juniors for growing to the next level.
JavaAWSMicroservicesKafkaRedis

Software Development Engineer 2, Amazon

Oct 2021 - Aug 2023Bengaluru, Karnataka, India

Software Engineer in the Developer Experience team, solving app certification and publishing problems.

  • Contributed to the design and Development of AutoFailure framework that is failure case agnostic to automate developer communication for App testing failures, helped to increase ZTP by ~6.5%.
  • Successfully reduced False Negatives of static testing by up to 12%, significantly minimizing the manual efforts required for validating apps.
  • Designed and implemented window manipulation for automatic testing of Android Apps on Windows, enabling efficient testing of approximately 200K apps using a unified framework.
  • Improved the user experience (UX) for the promotion portal on the App Store retail page and Developer portal, enhancing engagement and visibility.
  • Optimized queries and data retrieval time for App details by 50%, resulting in improved performance and response time.
  • Mentored an intern, providing guidance and support, and actively contributing to project planning in an Agile environment.
JavaAWSReactDynamoDBLambdaAPI Gateway

Backend Software Engineer - Secure Home Platform, McAfee

Dec 2018 - Sep 2021Bengaluru, Karnataka, India

Spearheaded a custom home security solution for an ISP partner, on Python, Cassandra, Redis, DynamoDB, Kafka, AWS, and Terraform.

  • Streamlined partner-specific changes by 50% through the implementation of a core solution that generalized custom modifications.
  • Resolved critical subscription failure cases, effectively mitigating DDOS-like attacks from routers and ensuring the security and stability of the system.
  • Designed and implemented a scalable custom feature-set module to seamlessly support various router models and versions, enabling broader compatibility and ease of integration.
  • Optimized the data flow design, resulting in a remarkable reduction of up to 40% in DB storage and network calls for the Router API, significantly improving efficiency and performance.
PythonCassandraRedisDynamoDBKafkaAWSTerraform

Android Software Engineer - Safe Connect (VPN client), McAfee

Oct 2017 - Dec 2018Bengaluru Area, India

Played a key role as a Software Engineer in Android, specialising in mobile app development for McAfee's Safe Connect, a security app providing secure network access through a VPN service.

  • Created robust VPN interfaces for the Android end-product, ensuring seamless integration and optimal user experience.
  • Developed a custom network state-machine to support Auto-start VPN and Activity-log features, enhancing functionality and user convenience.
  • Implemented support for Android 8 notification changes via SDK integration, ensuring compatibility with the latest platform updates.
AndroidJavaVPN ProtocolsNetwork Security

Backend Software Engineer - Apps scanner, McAfee

Jun 2016 - Oct 2017Bengaluru Area, India

Worked as a Software Engineer for the App Scanner team, utilizing technologies such as Python, Flask, XML, MySQL, and MongoDB.

  • Implemented optimizations that significantly increased the throughput of the app scanners, reducing repeat scanning of Android apps by up to 5000 per day.
  • Designed and implemented a robust pipe-filter service, employing pluggable codes, to efficiently process APK data.
  • Helped MySQL fix trigger bug: https://bugs.mysql.com/bug.php?id=87371
  • Collaborated with cross-functional teams to ensure seamless integration and optimal performance of the app scanning system.
PythonFlaskXMLMySQLMongoDB

Software Engineer Intern, Broadcom Limited

Jun 2015 - May 2016Bengaluru Area, India

Collaborated with the Bluetooth Software Development team, focusing on Bluetooth Stack for Android OS (AOSP).

  • Implemented and enhanced Bluetooth profiles, including AVRCP, MAP, and HID.
  • Successfully developed a service layer for AVRCP 1.6 CoverArt within a tight one-month development cycle.
  • Engineered a utility code that allowed testing of BLE-only APIs through a mobile app via reverse engineering of AOSP code.
  • Provided crucial support for the migration of the Bluetooth stack from Android N to Android M codebase.
AndroidBluetoothAOSP

Work

Side projects

Independent products I still run, on a low-touch cadence.

OneCue

Independent

Decision memory for coding agents. Surfaces the prior call, then stays silent.

  • CLI
  • Agent skills
  • TypeScript

AI SDK Patterns

Open source

Open-source TypeScript patterns, a composer, and a 16-chapter agent course.

  • AI SDK
  • Next.js
  • Open source

VibeFrames

Independent

Agentic Remotion workspace. A harness that edits a real composition.

  • Mastra
  • Remotion
  • TypeScript

Evercall

Independent

Voice agent that answers missed plumber calls, books the job, and texts the owner.

  • LiveKit
  • WebRTC
  • Python

Skills

The stack

Agent runtimes and voice systems now. Before that: SMS platforms, app certification, home-network security, Android.

Agent systems

Production runtimes: threads that keep state, tools that write, and a human stop before anything consequential.

  • Harnesses
  • Tool execution
  • Observational memory
  • Skills catalogs
  • Evals
  • Human-in-the-loop
  • Mastra
Voice

Low-latency duplex: VAD, barge-in, streaming STT/TTS, and tools that must not lie mid-call.

  • LiveKit
  • WebRTC
  • VAD
  • Barge-in
  • Streaming STT/TTS
Languages

The stack I actually ship in, across Salesforce, Amazon, and McAfee.

  • TypeScript
  • Python
  • Java
Frontend

Product surfaces for multi-turn agents and operator consoles.

  • React
  • Next.js
Backend

Services, queues, and the unglamorous path from request to durable write.

  • FastAPI
  • Node.js
  • Kafka
  • Redis
  • Microservices
Cloud & data

AWS-heavy production systems: messaging platforms, scanners, home-network backends.

  • AWS
  • Terraform
  • Lambda
  • DynamoDB
  • Cassandra
  • MySQL
  • MongoDB
Mobile

Earlier career: Android clients, VPN state machines, Bluetooth stack work.

  • Android
  • Java
  • AOSP

Writing

Notes from the work

WebRTC from First Principles: Why Voice Agents Run on UDP, Not HTTP

Essay · Aug 2026

Sound is a 20ms clock. TCP was built for documents. This is a ground-up tour of WebRTC — sampling, RTP, ICE, jitter buffers — and why every serious voice agent ends up as a media participant, not a server behind a POST.

The State Machine You Were About to Write Already Exists — LiveKit Voice Agents

Tutorial · Aug 2026

Phone agents need dates, room types, and callback numbers. The instinct is a CallState dataclass and next_question(). LiveKit Agents already solved that with AgentTask and TaskGroup — here's the hotel receptionist that shows how.

Tutorials Articles

Contact

Email me

The résumé is here. Email akashp1712@gmail.com.