# Mayank Pratap Singh
> Building high-performance production systems. Expert in distributed backends, scalable ML pipelines (RAG, MLOps), and autonomous robotics (ROS2, SLAM). Shipped systems for healthcare, aerospace, and autonomous navigation.

- Role: Full Stack, Robotics & ML Engineer
- Handle: @steeltroops
- Location: India
- Site: https://steeltroops.vercel.app
- Email: steeltroops.ai@gmail.com
- Profile JSON: https://steeltroops.vercel.app/person.json

## Summary
I build production systems. Full stack, ML pipelines, robotics. The kind of work where 'it mostly works' isn't good enough.
I've shipped across healthcare, aerospace, and autonomous systems. I pick hard problems, figure them out, and deliver something that actually holds up.
If it needs to move in the real world or scale in the cloud and survive production traffic, I've probably shipped it.

## About
I'm an engineer who builds across the full picture. Known across the community as @steeltroops, I work from frontend to backend to ML to hardware. I don't stop where my job title ends.
I've led teams, founded a startup, and worked in rooms where I was the least experienced person. The last one taught me the most.
I care about architecture. Not the buzzword kind. The kind where your system still works six months later when traffic doubles and nobody remembers how it was built.
I learn fast, build faster, and I'm honest about what I don't know. That last part saves more time than most people realize.
Most engineers pick a lane. I picked a problem space and learned whatever was needed to solve it. That's how you end up building ML pipelines on Monday and debugging ROS2 sensor nodes on Thursday.

## Experience
### Robotics C2 System Engineer @ Stealth
- Apr 2026 - Present
- Architected a multi-vehicle drone Command & Control ground station as the sole engineer: 12+ Rust and Go microservices, a React/CesiumJS operator interface, and a cross-platform Tauri desktop app spanning autopilot link through operator decision under a p95 sub-250ms telemetry-to-screen latency budget.
- Built the real-flight stack from the firmware layer up: custom MAVLink v2 libraries with cryptographic frame signing, pluggable transport adapters (serial/UDP/TCP/CRSF/DroneCAN), and PX4/ArduPilot drivers enabling signed arm, takeoff, mission, and return-to-launch paths verified on real autopilots and SITL.
- Implemented multi-vehicle swarm coordination with consensus-based task allocation (ACBBA), live formation flight along mission paths, and per-vehicle presence and safety state machines, surfaced through a 3D operator scene for planning, simulation, and live monitoring.
- Stood up a hardware-free test environment (Gazebo with PX4/ArduPilot SITL plus a custom fake-drone fleet simulator) and a signed, self-updating release pipeline across Windows, macOS, and Linux, including an offline field-kit build that runs the full stack on a single laptop with no cloud dependency.
- Defined the modular, plugin-based architecture through 280+ written decision records (ADRs/RFCs), keeping vehicle families, transports, simulators, and operator workflows as drop-in extensions rather than hard-coded paths.
- Tech: Rust, Go, React, CesiumJS, Tauri, MAVLink, PX4, ArduPilot, Gazebo, DroneCAN

### Software Engineer, Backend & Integration @ SS Innovations, India
- Oct 2024 - Nov 2025
- Designed multi-service architecture separating telemetry ingestion (Node.js), processing (Python/MLflow), and visualization (Next.js/PostgreSQL) layers to support 10x data volume growth with horizontal scaling capabilities.
- Led architecture decisions including schema design for time-series medical data, service boundary definitions, caching strategies with Redis, and observability pipelines using structured logging and CloudWatch integration.
- Built production ML pipelines for surgical anomaly detection with MLflow experiment tracking, achieving 99.2% uptime through comprehensive monitoring and rapid incident response during on-call rotations.
- Participated in production monitoring, logging and incident response to maintain continuous system availability.
- Tech: Node.js, Python, MLflow, Next.js, PostgreSQL, Redis, CloudWatch

### Research Intern - Flight Dynamics & Simulation @ Orbitx India Aerospace, India
- Jul 2023 - Dec 2023
- Designed trajectory optimization simulation systems using PPO and DDPG reinforcement learning models, improving algorithm convergence efficiency by 48% through modular architecture redesign and hyperparameter experimentation.
- Built structured experimentation pipelines with reproducible evaluation metrics and version-controlled models to support scalable aerospace research validation.
- Applied numerical methods, system modeling, and performance optimization to large-scale simulation workloads processing millions of trajectory calculations.
- Tech: Python, PPO, DDPG, Reinforcement Learning, Numerical Methods

### Founder & Tech Lead @ Inexia, India
- Oct 2022 - May 2024
- Led development of a VR-based research and simulation platform using Unreal Engine 5 for interactive experimentation, integrated with cloud dashboards built in Next.js and PostgreSQL to support experiment logging and real-time collaboration for 50+ users.
- Translated user feedback and product discovery discussions into technical feature implementations, shipping 8 validated iterations over 6 months in a fast-moving student startup environment.
- Designed modular backend services and structured data schemas for experiment tracking, while establishing documentation, version control workflows, and basic unit testing standards across a 10-member distributed student team.
- Tech: Unreal Engine 5, C++, Next.js, PostgreSQL, Unit Testing

### Vice President & Robotics Team lead @ Droid Robotics Club, GLAU
- Sep 2022 - Mar 2024
- Designed modular robotics architectures using ROS2, integrating perception, planning, and control stacks across distributed nodes with real-time pub/sub communication patterns.
- Implemented autonomous navigation pipelines including SLAM, RRT* path planning, sensor fusion (LiDAR/IMU), and state estimation for multi-sensor environments.
- Mentored 20+ engineering students on distributed systems debugging, ROS graph optimization, and software architecture best practices for production robotics.
- Tech: ROS2, SLAM, Path Planning, Sensor Fusion, Distributed Systems

## Projects
### Black Hole Simulation
Real-time relativistic ray-marching engine in the browser, solves Null Geodesic Equations in Kerr spacetime to render gravitational lensing, Doppler beaming, and volumetric accretion disk physics at interactive framerates. Rust physics kernel compiled to WASM; GLSL/WGSL shaders with WebGPU primary and WebGL 2.0 fallback.
- URL: https://blackhole-simulation.vercel.app/
- Source: https://github.com/steeltroops-ai/blackhole-simulation
- Tech: Next.js 14, WebGPU / WebGL 2.0, Rust + WASM, GLSL / WGSL, TypeScript, React Three Fiber, Bun

### OmniContext - Semantic Code Context Engine for AI Agents
Natively compiled Rust MCP server that exposes structured codebase abstraction to AI agents, AST parsing via tree-sitter across 16 languages, local semantic embeddings via CodeRankEmbed (ONNX runtime), hybrid SQLite index with HNSW vector search + BM25 full-text reranking. Zero cloud API calls.
- URL: https://omnicontextt.vercel.app/
- Source: https://github.com/steeltroops-ai/omnicontext
- Tech: Rust, TypeScript, MCP (Model Context Protocol), tree-sitter, ONNX Runtime, HNSW, SQLite, VS Code Extension

### MediLens - Clinical AI Intelligence Platform
Production multimodal AI diagnostics platform with 10 clinical modules across vision, audio, and structured data. Accuracy: RetinaScan 93.0%, ChestXplorer 97.8%, CardioPredict 99.8%, SpeechMD 95.2%, SkinSense 94.5%, Motor 93.5%, Cognitive 92.1%, NeuroScan 91.4%, RespiRate 93.2%, NRI Fusion 97.2%.
- URL: https://medilenss.vercel.app/
- Source: https://github.com/steeltroops-ai/MediLens
- Tech: Next.js 16, TypeScript, FastAPI, PyTorch, PostgreSQL, MLflow, Amazon Polly, Redis

### MedScribe AI - Agentic Clinical Documentation (MedGemma)
ReAct cognitive loop where MedGemma 27B reasons over clinical context and dispatches 5 Google HAI-DEF models as tools: MedASR, MedSigLIP-448 (image triage), MedGemma-4B (SOAP + ICD-10), TxGemma-2B (drug interactions), FHIR R4 compiler. 14s mean latency; 128ms deterministic fallback.
- URL: https://medscribbe.vercel.app/
- Source: https://github.com/steeltroops-ai/med-gemma
- Tech: Python 3.12, MedGemma 27B / 4B, TxGemma 2B, MedSigLIP-448, MedASR, LoRA Fine-tuning, FastAPI, Next.js 16, WebGPU, FHIR R4, HuggingFace Spaces, SSE

### ROS2 Robot Stack - Online Robotics Personal Lab
Full-stack multi-robot fleet management platform: ROS 2 Humble (Nav2, MoveIt2, Gazebo) bridged to the web via Fastify + rclnodejs + Socket.IO backend, with a Next.js 16 / React Three Fiber 3D dashboard rendering real-time telemetry and sensor overlays.
- URL: https://omniiverse.vercel.app/
- Source: https://github.com/steeltroops-ai/ros2-robot-stack
- Tech: ROS 2 Humble, Next.js 16, React Three Fiber, Fastify, Socket.IO, rclnodejs, FastAPI, PyTorch, YOLOv8, Nav2, MoveIt2, Gazebo, Docker

### NeuraForge - AI-Native Research Operating System
AI-agent-native research OS, orchestrates domain-specialized agents (math, coding, physics, biology, policy) in parallel, each maintaining long-context reasoning memory and persistent knowledge graphs. Turborepo monorepo: Next.js 14, Fastify + Socket.IO, PostgreSQL + Prisma, Clerk.
- URL: https://neuraaforge.vercel.app/
- Source: https://github.com/steeltroops-ai/NeuraForge
- Tech: Next.js 14, TypeScript, Fastify, Socket.IO, Turborepo, PostgreSQL, Prisma, Clerk, Bun

### My Portfolio - Agentic Full-Stack Personal Ecosystem
Production-grade personal ecosystem featuring an AI-automated blog generation engine with automated data collection, SEO optimization, and an integrated admin workspace for content orchestration.
- URL: https://steeltroops.vercel.app/
- Source: https://github.com/steeltroops-ai/my-portfolio-steeltroops
- Tech: Next.js 15, React 19, TypeScript, Tailwind CSS, Framer Motion, AI Block Generation, Admin Analytics, Bun

### RAG LLM Fact Checker
Production RAG system that indexes research papers and delivers fact-checked LLM responses with source attribution. Sub-200ms query latency via pgvector indexing, 512-token overlapping chunks, and embedding cache warming.
- URL: https://llm-fact-checker.vercel.app/
- Source: https://github.com/steeltroops-ai/llm-fact-checker
- Tech: Python, FastAPI, PostgreSQL, pgvector, LLM APIs, Vector Embeddings, Redis

### PaperForge-rs - Rust Semantic Research Retrieval Microservice
Production async Rust microservice for academic paper indexing and retrieval. Hexagonal architecture (Tokio, Axum) across a multi-crate workspace: gateway, search, ingestion, context, embedding-worker. Sea-ORM + raw SQLx for vector ops.
- URL: https://paperforge-rs.vercel.app/
- Source: https://github.com/steeltroops-ai/paperforge-rs
- Tech: Rust, Tokio, Axum, PostgreSQL, pgvector, Sea-ORM, HNSW, Prometheus, Docker, Next.js

### Metro-Quant - Adaptive Algorithmic Trading Bot (HackaTUM 2025)
Adaptive algorithmic trading bot built for the IMC Munich ETF Challenge at HackaTUM 2025, discovers alpha in live Munich city data (weather, air quality, flights) and trades on IMC's simulated exchange. Signal generation uses engineered features with Kalman filtering for noise reduction, producing [-1.0, 1.0] strength scores.
- Source: https://github.com/steeltroops-ai/Metro-Quant
- Tech: Python, Async I/O, Kalman Filter, Regime Detection, Streamlit, Hypothesis, Prometheus, Numba JIT

### ALYX - Distributed High-Energy Physics Analysis Orchestrator
Distributed analysis orchestrator for high-energy physics, processes collision events at petabyte scale, modeled on ALICE experiment workflows. Spring Boot 3.2 / Spring Cloud microservices: job scheduler, data router, result aggregator, ML resource optimizer, and real-time collaboration service.
- Source: https://github.com/steeltroops-ai/ALYX
- Tech: Java, Spring Boot 3.2, Spring Cloud, Apache Spark, Apache Kafka, React, TypeScript, Three.js, Monaco Editor, PostgreSQL

### CORE - Research-to-Commercialization Intelligence Platform
Hackathon platform (Track 1: Tech Transfer) that transforms research inputs into commercial intelligence, startup feasibility, VC investment summaries, and GTM strategy via AI pipelines.
- URL: https://core-five-phi.vercel.app/
- Source: https://github.com/steeltroops-ai/CORE
- Tech: Next.js, TypeScript, FastAPI, Python, LLM APIs, Render

### MemeCoin Casino - Provably Fair Solana DEX Casino
Decentralized provably fair casino on Solana (Anchor framework) supporting any SPL token, uses Solana blockhashes for transparent on-chain randomness with a complete proof trail per game.
- URL: https://memecoin-casino.vercel.app/
- Source: https://github.com/steeltroops-ai/memecoin-casino
- Tech: Rust, Solana, Anchor Framework, Next.js, TypeScript, Solana Wallet Adapter, Tailwind CSS

### Hiregeist - AI Job Placement Platform
Full-stack job placement platform on Next.js 15 / React 19 with a zero-trust security model, AI-powered candidate-to-role matching and real-time WebSocket communication.
- URL: https://hiregeist.vercel.app/
- Source: https://github.com/steeltroops-ai/Hiregeist
- Tech: Next.js 15, React 19, TypeScript, WebSockets, Zero-trust Auth

### AI-Powered Drug Research Platform
Collaborative computational drug research platform, AI-driven molecule generation and 2D protein folding visualization via Nvidia NeMo + RDKit cheminformatics pipelines.
- URL: https://drug-research.vercel.app/
- Tech: Next.js, TypeScript, MongoDB, Nvidia NeMo, RDKit, NextAuth, LLM APIs

### Research Knowledge Graph & Semantic Explorer
Structured knowledge graph mapping researchers, publications, and datasets, 2,000+ papers indexed with embedding-based semantic search, pgvector indexing, and retrieval reranking.
- Tech: TypeScript, PostgreSQL, pgvector, LLM APIs, Vector Embeddings, FastAPI

### Niryo LLM Robotic Arm Integration
Niryo Ned2 physical arm controlled via natural language, LLM interprets intent, plans pick-and-place sequences, executes through ROS2 action servers in real time.
- Source: https://github.com/steeltroops-ai/Niryo-llm-robo
- Tech: Python, ROS2, Computer Vision, LLM APIs, Niryo SDK, OpenCV

### Robot Bionic Arm
5-DOF robotic arm from first principles, custom PID per joint, IK solver for Cartesian task control, ROS2 sensor fusion from joint encoders and force sensors at 100Hz.
- Tech: ROS2, C++, Python, PID Control, Inverse Kinematics, 3D Printing, Sensor Fusion

### Maze-Bot - ROS2 Autonomous Navigation Platform
ROS2 autonomous navigation research platform integrating SLAM, multi-algorithm path planning, and a clean benchmarking framework.
- Tech: ROS2, C++, SLAM, Nav2, A* / RRT / DWA, LiDAR, Gazebo

### VR Firefighting & Flood Training Simulator
Physics-based VR training simulator in Unreal Engine 5 for firefighter and flood emergency response, dynamic fire spread, water physics, structural response, and procedurally generated scenarios. Built at Inexia.
- Tech: Unreal Engine 5, C++, Chaos Physics, VR (OpenXR), Blueprints

### Black Hole Simulation — Native C++ / OpenGL Engine
Native desktop black hole renderer in C++17 / OpenGL 3.3, Schwarzschild metric, geodesic motion, real-time gravitational lensing. Standalone desktop counterpart to the WebGPU browser simulation.
- Source: https://github.com/steeltroops-ai/blackhole-sim
- Tech: C++17, OpenGL 3.3, GLFW, GLAD, CMake, GLSL

### DebrisGuard — AI Space Debris Tracking System
AI-powered space debris tracking system, TLE/SGP4 ingestion, orbital trajectory simulation, and ML-based collision probability scoring for active satellite operators.
- Source: https://github.com/steeltroops-ai/DebrisGuard
- Tech: Python, JavaScript, Orbital Mechanics, TLE / SGP4, Kubernetes, AI Trajectory Prediction

### PyTorch Transformer — Attention Is All You Need
Transformer architecture from Vaswani et al. 2017 implemented from scratch, multi-head self-attention, positional encoding, encoder-decoder stacks, layer normalization, beam search decoding.
- Source: https://github.com/steeltroops-ai/Pytorch-Transformer
- Tech: Python, PyTorch, Transformers, NLP, Jupyter Notebook, Matplotlib

### MDC Dataset Citation Classification
Ensemble ML system classifying dataset citations in scientific literature, SciBERT contextual embeddings combined with gradient boosting and SVM classifiers, trained on annotated papers.
- Tech: Python, SciBERT, Scikit-learn, Ensemble Methods, NLP, Hugging Face

### Eidolon - AI-Powered Knowledge Graph
Personal knowledge management system that visualizes ideas as an interactive force-directed graph, nodes linked by semantic relationships, Gemini 2.5 Flash AI assistant for contextual queries. Next.js 16, React 19, TypeScript 5, Bun, Tailwind CSS 4 glass-morphism.
- URL: https://eidolon-two.vercel.app/
- Source: https://github.com/steeltroops-ai/Eidolon
- Tech: Next.js 16, React 19, TypeScript, Google Gemini 2.5, Tailwind CSS 4, Bun, Jest

### GreenTwin - AI Carbon Footprint Digital Twin
AI carbon footprint digital twin, Gemini 1.5 Pro coaching with predictive emission forecasting and smart intervention nudges. Passive tracking via Chrome MV3 extension (background service worker across Amazon, Kayak, news sites). Hackathon project.
- URL: https://green-twin.vercel.app/
- Source: https://github.com/steeltroops-ai/GreenTwin
- Tech: Next.js 15, TypeScript, Google Gemini 1.5 Pro, Chrome Extension (MV3), Clerk, WebSocket, Bun

### FlowBot - Visual Chatbot Flow Builder
Professional visual flow builder for designing conversational chatbot flows, drag-and-drop canvas with 12+ node types (messages, logic, input, integration, utility, action) and smooth animations.
- URL: https://flowbott.vercel.app/
- Source: https://github.com/steeltroops-ai/FlowBot
- Tech: React 18, TypeScript, ReactFlow, Zustand, Tailwind CSS, Vite

### Finance Management App
Full-stack finance platform with Plaid multi-bank account linking, real-time balance tracking, Dwolla inter-account transfers, and categorized spending analytics. SSR authentication with Appwrite BaaS.
- URL: https://banking-jet.vercel.app/sign-in
- Source: https://github.com/steeltroops-ai/finanace-management-banking-app
- Tech: Next.js, TypeScript, Plaid API, Dwolla, Appwrite, SSR Auth

### Neza Digital Marketplace
B2C service marketplace connecting households with verified local providers, multi-role architecture (consumer, provider, admin), real-time availability management, Razorpay payment gateway with webhook confirmation.
- Source: https://github.com/steeltroops-ai/Neza
- Tech: Next.js, TypeScript, Node.js, Razorpay, RBAC, PostgreSQL

### House Price Prediction - End-to-End MLOps Pipeline
End-to-end ML pipeline with ZenML orchestration and MLflow experiment tracking, data ingestion, schema validation, feature engineering, training, and automated deployment gating.
- Source: https://github.com/steeltroops-ai/House-Price-Prediction-with-MLOps
- Tech: Python, Scikit-learn, ZenML, MLflow, Pandas, Feature Engineering

## Social
- twitter: https://x.com/steeltroops_ai
- github: https://github.com/steeltroops-ai
- instagram: https://instagram.com/steeltroops_ai
- linkedin: https://linkedin.com/in/steeltroops-ai
- bento: https://own.page/steeltroops

## Education
- B.Tech, Computer Science (AI & ML), GLA University

## Optional
- [Full HTML site](https://steeltroops.vercel.app): interactive portfolio
- [Structured person graph](https://steeltroops.vercel.app/person.json): schema.org Person + experience + projects