Building things · Learning in public
Akshat Kharkwal
Student. I build small tools — mostly machine learning, mostly to solve problems I have.
Writing
What I Learned Trying to Predict Kanpur's Air Quality
→
Next-day AQI forecasting with real CPCB data — and why 67% accuracy taught me more than 90% would have.
Sep 2026
7 min read
Projects
Antigravity Studio
→
Modern iPadOS & Desktop agentic workspace. Features 24 God plugins, specialized subagents fleet, MCP connectors, real-time code runner, and background automations engine.
PWA
TypeScript
Gemini API
MCP
Automations
donecheck
→
Adaptive verification engine and Claude Code skill. Evaluates implementation changes against test suites and generates tailored verification reports based on developer experience level.
Developer Tools
Claude Code
Verification
Testing
JARVIS
→
Autonomous AI agent with 55+ tools spanning browser automation, screen vision, RAG, and sandboxed code execution. Writes, tests, and saves its own tools at runtime. Includes an Experiment Mode that turns research questions into reproducible experiments.
Python
LLMs
Agents
Self-extension
Singapore Island Biodiversity
→
ML pipeline investigating vegetation change across six Singapore micro-islands. Combines multi-temporal Sentinel-2 satellite imagery with 22,000+ iNaturalist observations. Evaluates XGBoost against persistence and linear-trend baselines.
Python
XGBoost
Remote sensing
Time series
Weather-Predictor
→
Random Forest classifier trained on 105k real CPCB observations. Chronological split, persistence baseline, 67.1% accuracy.
Python
scikit-learn
pandas
JEE-Quiz-Studio
→
Zero-dependency quiz app with per-topic accuracy tracking. Built to make my own revision more targeted.
JavaScript
localStorage
GitHub Pages
Assistive Stick
→
Arduino walking stick for visually impaired navigation. Ultrasonic obstacle detection with distance-scaled beep frequency. Team project — I wrote the firmware.
Arduino
C++
Sensors