Akhilesh Chandra Machine learning engineer

I build the data and modelsthat physical AI runs on.

Machine learning engineer working on perception pipelines for robotics: segmentation, 3D mesh, pose and PII redaction across egocentric, stereo and fisheye video. Before that I built synthetic data pipelines for enterprise agents, shipped a voice assistant to npm and spent three years finding critical vulnerabilities in software people use every day.

01

Experience

MJAK LabsSeptember 2026 - PresentPhysical AI and Robotics

Founder

  • Evals: building evaluations and benchmarks for 3D hand pose, gloves, depth, mesh and segmentation models. Also building our own 3D hand pose model.
  • Robotruth: a CLI tool that checks whether a change in a robot policy makes a real difference before deploying it to hardware.
  • NVIDIA GR00T: fine tuned and evaluated a GR00T policy on an NVIDIA A40. Our V3 checkpoint achieved 92 successes across 100 held out MuJoCo trials at approximately 30 Hz with zero policy errors, infrastructure errors or joint limit corrections.
  • Robot API: building APIs to run YAM, SO-100 and SO-101 arms remotely.
  • Open source hardware: UMI Gripper and a robotic arm coming soon.
  • Running the Physical AI Fellowship backed by SFCompute for students and researchers working on open source robotics. Disbursed $4.5k in compute.
Evaluation and BenchmarkingRobot PolicyRL Post Training
Deccan AIJune 2026 - July 2026Synthetic Data

Machine Learning Engineer

  • Built a complete ML pipeline that generates synthetic databases across 10 tools to train enterprise agents.
  • Reached an 89% unique and related data fill rate across 60k rows.
  • Built approaches with and without LLMs for the Data forge pipeline. Also built an approach that compares changes across multiple commits.
  • Designed the architecture for the chain of thoughts project.
Synthetic DataLLM pipelinesAgents
Activate VCJune 2026 - PresentFellowship

AI Fellow

  • Joined as an AI Fellow to learn and collaborate with other fellows on LLMs and AI agents.
Hebbian Robotics (YC S26)September 2026Open Source Contributions

Open Source Contributor

  • Passed channel topics down to the MCAP reader so selecting a channel by ID no longer decompresses every stream in the file. Reading a small state channel now skips unrelated chunks even when the file contains gigabytes of camera data. Filtering by channel ID also works when several channels share a topic.
  • Stopped SQL errors echoing the internal preview wrapper. A failed query was showing people a DESCRIBE SELECT * FROM (...) statement they never wrote. Errors during execution now omit that location block and keep the diagnostic text. Parse errors retain it because they refer to the caller's own SQL.
MCAPDuckDBPythonRobotics data
Peak AI RoboticsJuly 2026 - August 2026Robotics

Machine Learning Engineer

  • Built the pipeline for segmentation and PII redaction on egocentric, stereo and fisheye video.
  • Wired up GCP so videos could be sourced and a GPU job started in one click.
GCPGPU pipelinesComputer Vision
Stealth StartupAugust 2026 - PresentMachine Learning

Machine Learning Engineer

  • Built the pipeline that generates segmentation, 3D mesh, masks, gloves and PII redaction for egocentric, stereo and fisheye video.
Segmentation3D meshPII redactionEgocentric video
Asai LabsApril 2026 - June 2026Computer Vision

Computer Vision Engineer

  • Created the architecture and defined every approach from scratch to build the shelf fill rate detection model for 180 stores of the biggest bakery in the CEE region.
Retail visionShelf analyticsModel architecture
IndependentAugust 2023 - October 2026Security

Security Researcher

  • Found 8 critical vulnerabilities, including remote code execution, LFI and broken access control at Meta.
  • Reported a critical denial of service in Sequoia Open PGP, session hijacking in Firefox Nightly and path traversal in Cash App and Expo Go.
  • Reported usability bugs at Booking.com and YouTube.
RCEAccess controlResponsible disclosure
GoAhead Ventures, Expert Dojo and LvlUp VenturesJanuary 2024 - January 2025Venture

Venture Scout

  • Used my network to source early stage startups across AI, e-commerce and fintech.
  • Evaluated founding teams and product architecture for investment potential.
  • Built a working sense of what makes a startup scalable, what investors look for and what users need.
GoldNestAugust 2023 - March 2024Fintech

Founder

  • GoldNest set out to make gold investing accessible to everyone: digital gold, physical gold, bonds and ETFs.
  • Built the prototype and presented it at IIT Roorkee and IIM Lucknow.
AxtraMay 2022 - August 2023E-commerce

Co-Founder

  • An e-commerce startup focused on Gen Z fashion. In a team of 5 we built the app and brought brands on as vendors.
  • Led marketing that reached 11.1k+ people and 500+ followers in 3 days.
02

Projects

Dataset for Physical AIDatasets and Models

Physical AI

Ready to train logs and usable models for physical AI in dark stores, factories and warehouses.

  • Human pose detection: identifies 17 human joints and the resulting pose in real time at 95%+ accuracy.
  • Object detection v1: trained with just 5 validation images. Identifies household objects instantly in real time at 80%+ accuracy.
  • Object detection v2: trained on 110k SKU images. Identifies objects of any size in any environment at 95%+ accuracy.
  • Human and object detection: trained on RF-DETR. Identifies both in real time. Converted to ONNX runtime and packaged as an Android APK that performs equally well on device.
RF-DETRONNXAndroidPose estimation
StablePeStablecoins

Stablecoin Payments

A stablecoin payments app built for B2B businesses and remittance with a waitlist of businesses and individuals.

  • AML Relay API: an automated safety tool that checks a receiver's wallet against official government blocklists such as the OFAC list, stopping illegal payments before they happen.
  • A verification system that reads a wallet's on-chain transaction history to compute a risk score, automatically blocking wallets above 85 and flagging a warning for the rest.
  • Scan dashboard: real time monitoring for researchers and analysts, tracking whale transactions, wallet histories and live activity across the top stablecoins and chains.
BlockchainAMLRisk scoring
SpectraDesktop App

One Tool for Every Kind of Bug Hunting

  • A Windows app built with Electron and React that brings asset discovery, traffic interception and vulnerability scanning into one place. It replaces several separate CLI tools.
  • A custom MITM proxy in Node.js that intercepts and modifies live HTTPS traffic, using SQLite to handle large request volumes without slowing down.
  • Includes Intercept, Repeater, Intruder and History. Also offers free collaborator URLs, disposable mail and a VPN.
  • A full xterm.js terminal with WSL support so custom Python or Go scripts run inside the app.
ElectronReactNode.jsSQLite
Voice OSnpm Package

Multilingual Voice Assistant

  • A cross platform AI agent in an npm package that automates tasks by voice, supporting 8 languages across the STT and TTS pipelines. 1100+ downloads.
  • Integrated STT (Vosk), TTS (Edge) and an LLM pipeline so the whole loop runs on the user's device.
  • Added document processing, music control through Spotify and task automation.
VoskEdge TTSLLMnpm
Speech to TextReal Time

Self Hosted Streaming Service

  • Streaming speech to text with Vosk and FastAPI, exposed over WebSocket for real time transcription.
  • Processes 16-bit PCM mono audio at 16kHz straight from the microphone without explicit buffering on the server.
  • Streams partial and final transcriptions so feedback arrives in near real time during live speech.
  • Supports both WebSocket streaming and HTTP file transcription. Uses FFmpeg and a fallback decoder to handle different codecs.
VoskFastAPIWebSocketFFmpeg
Text to SpeechSynthesis

Speech Synthesis

  • A text to speech model written in Python using Microsoft Edge voices.
PythonEdge TTS
03

Education

University of LucknowOctober 2021 - June 2024Lucknow

Bachelor of Science in Computer Science

Major subjects: Computer Science, Maths and Physics.

Rani Laxmi Bai Memorial SchoolApril 2020 - April 2021Lucknow

Intermediate

Percentage: 85.6% out of 100.

Physics, Chemistry, Maths and Computer Science.

Rani Laxmi Bai Memorial SchoolApril 2018 - April 2019Lucknow

High School

Percentage: 80% out of 100.

Physics, Chemistry, Maths and Computer Science.

04

Skills and Interests

Programming

JavaScript, Python, React, HTML, CSS, SQL, Linux, APIs, Node.js, Git and MongoDB.

Languages

Hindi, English, Awadhi and Bhojpuri.

Interests

Chess, football and the internet.

Writing

Writing something.