I Ship
Production
Systems

location: NYC // status: compiling // nyu '26 // gpa: 0.00
late night coding session? same.
scroll

Hackathons

06 HACKATHONS

01/Karen

Karen — photo 1
Karen — photo 2
IMG_REF: 01/02

Karen gets results.

10
Real channels
4
Personalities
🦞
Karen gets results

Karen is a mischievous AI agent with one purpose: follow up until the matter is resolved. She escalates through 10 real integrated channels — email, SMS, WhatsApp + voice call, OSINT research, email CC, Slack, Discord @everyone, a live GitHub commit to a public accountability page, Google Calendar seizure, and a FedEx overnight letter — each with a different personality mode (Passive Aggressive, Corporate, Genuinely Concerned, Life Coach). The Circle web app lets anyone in your group trigger Karen against anyone else. The gamified view renders the escalation as a pixel art RPG with Karen as a crab sprite hunting the target through a village.

ROLEFull-stack solo build: OpenClaw skill architecture, FastAPI backend, all 10 channel integrations (Gmail SMTP, Twilio SMS + Voice, WhatsApp Business API, Slack Bot API, Discord Bot, Google Calendar API, GitHub Contents API, FedEx Rate + Ship API), SSE streaming, ElevenLabs TTS demo intro, personality system generating all message variants dynamically.
OpenClawFastAPINext.js 15TwilioElevenLabsDiscord BotSlack APIGoogle Calendar APIGitHub APIFedEx APISSE Streaming
MischiefClaw Hackathon (AI Tinkerers NYC / betaworks)PROJECT_1 OF 6

Projects

27 PROJECTS
$ gh repo list --source --no-forks
SYNCED · 27 REPOS
PUBLISHED ON NPM

MarkPush

Push markdown from your terminal to your iPhone. One command, instant delivery. End-to-end encrypted with AES-256-GCM, dual transport (WiFi via mDNS + Supabase cloud relay), and an MCP server so Claude can push directly to your phone.

GoSwiftTypeScriptSwiftUITCASupabaseMCP
Swift
76% → 2% → 61% ACCURACY

Deep-Learning-Project-3

FGSM, PGD, I-FGSM attacks on ResNet-34. PGD reduced Top-1 accuracy from 76% to 2%. Adversarial training restored it to 61%.

PyTorchResNet-34DenseNet-121ImageNet-1K
Jupyter Notebook
5K+ TRACKS, 96% MACRO-F1

SuperPlay

AI-powered Spotify playlist curator. ETL from Spotify/Genius APIs, transformer-based lyric embeddings, XGBoost classifier, Flask API, Docker deployment.

PythonXGBoostHuggingFaceFlaskDocker
Python

postmortem

A decision-archaeology agent for any codebase. Built with Claude Opus 4.7

TypeScript

brain-explorer

3D knowledge graph explorer for 375+ personal wiki articles. Renders semantic embedding clusters as glowing nodes in a nebula-themed WebGL scene using react-three-fiber, with bloom postprocessing, arc camera fly-ins, a command palette, and glassmorphic article panels.

TypeScript

brain

Self-improving personal knowledge base — Claude Code + Obsidian, local-first RAG

HTML

gaze-diffuse

Steering text readability in masked diffusion LLMs using eye-tracking gaze signals — applied to all token positions simultaneously at each denoising step.

Python
1

ouroboros

Self-improving agent engine. Start it, walk away, come back to find it improved itself.

Python

karen

Karen is a mischievous OpenClaw-powered follow-up agent built for the MischiefClaw Hackathon. She treats every non-response as a crisis and every crisis as an opportunity to escalate — across 10 channels, with 4 distinct personalities, against anyone in The Circle.

TypeScript

askNYC

Voice + camera AI agent for NYC. Point your phone at any restaurant, building, or construction site, ask a question, and it answers using real city data , all spoken back in real time. Gemini sees the camera, queries 5 NYC Open Data APIs, and the dashboard lights up with data cards and map pins.

TypeScript

Career-Datacenter

Python

Flock

Flock is an AI group travel planning app.

TypeScript

raag

TypeScript

system-design-mastery

Master system design from basics to interview-level — interactive, visual, free.

JavaScript

imessage-scheduler

iMessage message scheduler with FIFO queue and macOS gateway

TypeScript

Coro

JavaScript
1
OPEN-SOURCE TOOL

code-rot

Open-source spaced-repetition engine for interview prep. Recommendation engine for problems, adaptive scheduling based on forgetting curves, and progress visualization.

PythonSpaced RepetitionOpen Source
Python

HPML-Project

Python

MarketPulse-AI

MarketPulse AI is a real-time market insight chatbot. It continuously ingests live streams of financial data – such as breaking news headlines and price movements – and uses Retrieval-Augmented Generation (RAG) to answer traders’ and analysts’ questions with up-to-the-second context.

Python
2

Our-Scrabble

Python

Orion

Orion Veil is a web-first, dark-sci-fi text-adventure powered by OpenAI function-calling. Chat with an in-world AI ally, upload images or files as clues, and crack ARG-style ciphers and steganography puzzles to outwit a rogue super-intelligence— all in a 2-3 hour, fully-voice-ready narrative.

TypeScript
85% DETECTION ACCURACY

Garden-Path-Parser

Self-healing NLU toolkit with 20+ ambiguity templates, BERT-based semantic scoring for reranking. 85% detection accuracy. Packaged as pip module.

PythonBERTNLPpip
Python

Scraper-Project

A robust web scraper for extracting hotel marketing information from hotel websites. This tool extracts basic hotel information, images, venues, and nearby attractions using modern web scraping techniques combined with AI for enhanced data extraction.

Python

rahilsinghi

Config files for my GitHub profile.

WindProject

Wind Power Prediction using Machine Learning. Used the wind power data from a plant in India.

Python

DL_PROJECT_1

modified residual network (ResNet) architecture with the highest test accuracy on the CIFAR-10 dataset, under 5 million parameters with a validation accuracy of 95.6%

Jupyter Notebook

DALL-E-Clone

This project is a clone of the DALL-E image generation application, allowing users to input prompts and receive generated images using OpenAI's DALL-E API.

PUBLISHED ON NPM

MarkPush

Push markdown from your terminal to your iPhone. One command, instant delivery. End-to-end encrypted with AES-256-GCM, dual transport (WiFi via mDNS + Supabase cloud relay), and an MCP server so Claude can push directly to your phone.

GoSwiftTypeScriptSwiftUITCASupabaseMCP
Swift
76% → 2% → 61% ACCURACY

Deep-Learning-Project-3

FGSM, PGD, I-FGSM attacks on ResNet-34. PGD reduced Top-1 accuracy from 76% to 2%. Adversarial training restored it to 61%.

PyTorchResNet-34DenseNet-121ImageNet-1K
Jupyter Notebook
5K+ TRACKS, 96% MACRO-F1

SuperPlay

AI-powered Spotify playlist curator. ETL from Spotify/Genius APIs, transformer-based lyric embeddings, XGBoost classifier, Flask API, Docker deployment.

PythonXGBoostHuggingFaceFlaskDocker
Python

postmortem

A decision-archaeology agent for any codebase. Built with Claude Opus 4.7

TypeScript

brain-explorer

3D knowledge graph explorer for 375+ personal wiki articles. Renders semantic embedding clusters as glowing nodes in a nebula-themed WebGL scene using react-three-fiber, with bloom postprocessing, arc camera fly-ins, a command palette, and glassmorphic article panels.

TypeScript

brain

Self-improving personal knowledge base — Claude Code + Obsidian, local-first RAG

HTML

gaze-diffuse

Steering text readability in masked diffusion LLMs using eye-tracking gaze signals — applied to all token positions simultaneously at each denoising step.

Python
1

ouroboros

Self-improving agent engine. Start it, walk away, come back to find it improved itself.

Python

karen

Karen is a mischievous OpenClaw-powered follow-up agent built for the MischiefClaw Hackathon. She treats every non-response as a crisis and every crisis as an opportunity to escalate — across 10 channels, with 4 distinct personalities, against anyone in The Circle.

TypeScript

askNYC

Voice + camera AI agent for NYC. Point your phone at any restaurant, building, or construction site, ask a question, and it answers using real city data , all spoken back in real time. Gemini sees the camera, queries 5 NYC Open Data APIs, and the dashboard lights up with data cards and map pins.

TypeScript

Career-Datacenter

Python

Flock

Flock is an AI group travel planning app.

TypeScript

raag

TypeScript

system-design-mastery

Master system design from basics to interview-level — interactive, visual, free.

JavaScript

imessage-scheduler

iMessage message scheduler with FIFO queue and macOS gateway

TypeScript

Coro

JavaScript
1
OPEN-SOURCE TOOL

code-rot

Open-source spaced-repetition engine for interview prep. Recommendation engine for problems, adaptive scheduling based on forgetting curves, and progress visualization.

PythonSpaced RepetitionOpen Source
Python

HPML-Project

Python

MarketPulse-AI

MarketPulse AI is a real-time market insight chatbot. It continuously ingests live streams of financial data – such as breaking news headlines and price movements – and uses Retrieval-Augmented Generation (RAG) to answer traders’ and analysts’ questions with up-to-the-second context.

Python
2

Our-Scrabble

Python

Orion

Orion Veil is a web-first, dark-sci-fi text-adventure powered by OpenAI function-calling. Chat with an in-world AI ally, upload images or files as clues, and crack ARG-style ciphers and steganography puzzles to outwit a rogue super-intelligence— all in a 2-3 hour, fully-voice-ready narrative.

TypeScript
85% DETECTION ACCURACY

Garden-Path-Parser

Self-healing NLU toolkit with 20+ ambiguity templates, BERT-based semantic scoring for reranking. 85% detection accuracy. Packaged as pip module.

PythonBERTNLPpip
Python

Scraper-Project

A robust web scraper for extracting hotel marketing information from hotel websites. This tool extracts basic hotel information, images, venues, and nearby attractions using modern web scraping techniques combined with AI for enhanced data extraction.

Python

rahilsinghi

Config files for my GitHub profile.

WindProject

Wind Power Prediction using Machine Learning. Used the wind power data from a plant in India.

Python

DL_PROJECT_1

modified residual network (ResNet) architecture with the highest test accuracy on the CIFAR-10 dataset, under 5 million parameters with a validation accuracy of 95.6%

Jupyter Notebook

DALL-E-Clone

This project is a clone of the DALL-E image generation application, allowing users to input prompts and receive generated images using OpenAI's DALL-E API.

PUBLISHED ON NPM

MarkPush

Push markdown from your terminal to your iPhone. One command, instant delivery. End-to-end encrypted with AES-256-GCM, dual transport (WiFi via mDNS + Supabase cloud relay), and an MCP server so Claude can push directly to your phone.

GoSwiftTypeScriptSwiftUITCASupabaseMCP
Swift
76% → 2% → 61% ACCURACY

Deep-Learning-Project-3

FGSM, PGD, I-FGSM attacks on ResNet-34. PGD reduced Top-1 accuracy from 76% to 2%. Adversarial training restored it to 61%.

PyTorchResNet-34DenseNet-121ImageNet-1K
Jupyter Notebook
5K+ TRACKS, 96% MACRO-F1

SuperPlay

AI-powered Spotify playlist curator. ETL from Spotify/Genius APIs, transformer-based lyric embeddings, XGBoost classifier, Flask API, Docker deployment.

PythonXGBoostHuggingFaceFlaskDocker
Python

postmortem

A decision-archaeology agent for any codebase. Built with Claude Opus 4.7

TypeScript

brain-explorer

3D knowledge graph explorer for 375+ personal wiki articles. Renders semantic embedding clusters as glowing nodes in a nebula-themed WebGL scene using react-three-fiber, with bloom postprocessing, arc camera fly-ins, a command palette, and glassmorphic article panels.

TypeScript

brain

Self-improving personal knowledge base — Claude Code + Obsidian, local-first RAG

HTML

gaze-diffuse

Steering text readability in masked diffusion LLMs using eye-tracking gaze signals — applied to all token positions simultaneously at each denoising step.

Python
1

ouroboros

Self-improving agent engine. Start it, walk away, come back to find it improved itself.

Python

karen

Karen is a mischievous OpenClaw-powered follow-up agent built for the MischiefClaw Hackathon. She treats every non-response as a crisis and every crisis as an opportunity to escalate — across 10 channels, with 4 distinct personalities, against anyone in The Circle.

TypeScript

askNYC

Voice + camera AI agent for NYC. Point your phone at any restaurant, building, or construction site, ask a question, and it answers using real city data , all spoken back in real time. Gemini sees the camera, queries 5 NYC Open Data APIs, and the dashboard lights up with data cards and map pins.

TypeScript

Career-Datacenter

Python

Flock

Flock is an AI group travel planning app.

TypeScript

raag

TypeScript

system-design-mastery

Master system design from basics to interview-level — interactive, visual, free.

JavaScript

imessage-scheduler

iMessage message scheduler with FIFO queue and macOS gateway

TypeScript

Coro

JavaScript
1
OPEN-SOURCE TOOL

code-rot

Open-source spaced-repetition engine for interview prep. Recommendation engine for problems, adaptive scheduling based on forgetting curves, and progress visualization.

PythonSpaced RepetitionOpen Source
Python

HPML-Project

Python

MarketPulse-AI

MarketPulse AI is a real-time market insight chatbot. It continuously ingests live streams of financial data – such as breaking news headlines and price movements – and uses Retrieval-Augmented Generation (RAG) to answer traders’ and analysts’ questions with up-to-the-second context.

Python
2

Our-Scrabble

Python

Orion

Orion Veil is a web-first, dark-sci-fi text-adventure powered by OpenAI function-calling. Chat with an in-world AI ally, upload images or files as clues, and crack ARG-style ciphers and steganography puzzles to outwit a rogue super-intelligence— all in a 2-3 hour, fully-voice-ready narrative.

TypeScript
85% DETECTION ACCURACY

Garden-Path-Parser

Self-healing NLU toolkit with 20+ ambiguity templates, BERT-based semantic scoring for reranking. 85% detection accuracy. Packaged as pip module.

PythonBERTNLPpip
Python

Scraper-Project

A robust web scraper for extracting hotel marketing information from hotel websites. This tool extracts basic hotel information, images, venues, and nearby attractions using modern web scraping techniques combined with AI for enhanced data extraction.

Python

rahilsinghi

Config files for my GitHub profile.

WindProject

Wind Power Prediction using Machine Learning. Used the wind power data from a plant in India.

Python

DL_PROJECT_1

modified residual network (ResNet) architecture with the highest test accuracy on the CIFAR-10 dataset, under 5 million parameters with a validation accuracy of 95.6%

Jupyter Notebook

DALL-E-Clone

This project is a clone of the DALL-E image generation application, allowing users to input prompts and receive generated images using OpenAI's DALL-E API.

SPEED

Experience

WORK HISTORY
01Kismet
May 2025 – Jul 2025

AI / Data Engineering Intern

New York, NY

Hospitality tech startup building tools for hotels: scraping, booking integrations, tracking, and analytics.

FastAPINestJSGPT-4oGeminiMCPRedisAlloyDBGCP
02Kismet
Sept 2025 – Dec 2025

Computer Engineering Intern — Tracking Engineer

New York, NY

Backend APIs, analytics infrastructure, and cross-domain measurement for hotel marketing.

BigQueryGA4GTMPostgreSQLPython
03Krushal Open Commerce
Aug 2023 – Mar 2024

Software Engineering Intern

Chennai, India

AgriTech startup building apps for dairy operations and veterinary workflows.

ReactAPI DesignProduct Management

Education

ACADEMIC

NYU Tandon School of Engineering

Master of Science in Computer Engineering

2024 – 2026New York, NY
GPA0.00/ 4.0
Deep LearningNatural Language UnderstandingEfficient AI & Hardware Accelerator DesignDesign & Analysis of Algorithms

Edge AI Research — OptiReality VIP under Prof. Sai Qian Zhang

SSN College of Engineering

Bachelor of Engineering in Computer Science

2019 – 2023Chennai, India
GPA0.0/ 10.0
Data StructuresOperating SystemsComputer NetworksMachine Learning

Led 4-person engineering team at Krushal during final year

Skills

Loading graph...

Recent

10 POSTS

About

THE PERSON
Rahil Singhi

AI and Software Engineer building agentic workflows, backend systems, and data pipelines at the intersection of practical engineering and AI research. Currently finishing my MS in Computer Engineering at NYU (3.88 GPA), where I split time between edge AI research and shipping production code at startups.

>Ship production AI systems
>Build agentic tool-calling pipelines
>Design distributed systems
>Write clear documentation (yes, really)

Also writes literary fiction → @lostinpages_lp

// don't be a stranger

SayHi_

Looking for an AI + Software Engineer who ships? Let's talk.

Download Resume

Role-specific versions available on request.

PYTHONGOTYPESCRIPTPYTORCHFASTAPIgRPCDISTRIBUTED SYSTEMSLLM PIPELINESMCP TOOLSEDGE AIREACTDOCKERAWSGCPRAGFAISSGEMINIWEBSOCKETSREDISPOSTGRESQLBIGQUERYAGENTIC AIQUANTIZATIONHUGGINGFACEPYTHONGOTYPESCRIPTPYTORCHFASTAPIgRPCDISTRIBUTED SYSTEMSLLM PIPELINESMCP TOOLSEDGE AIREACTDOCKERAWSGCPRAGFAISSGEMINIWEBSOCKETSREDISPOSTGRESQLBIGQUERYAGENTIC AIQUANTIZATIONHUGGINGFACE