Loading the scan (12 MB)

Bengaluru, India

Akash Ainapur.
This is my desk.

I scanned it in 3D. Everything I have built started here, so the tour of my work is a tour of this table: robots, ML systems and films.

Scroll, or press → to walk through it.

01 · Selected work

Five things that run.

Each one names what I built, what a teammate built, and what is still unproven. The labels are deliberate.

  1. 01
    Prize winnerTeam project

    KRISHIBODH: a gantry robot that farms a plant bed

    A 4-axis Cartesian robot that drives to each plant, photographs it, checks the soil and waters only when it is dry.

    • 4 axesSource: Stepper setup in sketch_nov8a.ino; checked in the résumé line audit of 13 Oct 2026
    • 6 serial commandsSource: Command handlers in sketch_nov8a.ino
    • 96 captures, 13 runsSource: Counted *_raw.jpg files and demo_runs folders in the repo

    C++ (Arduino) · Python · Flask · OpenCV · Gemini API

  2. 02
    Team projectPaper under review

    AI-Driven Auto Deployment Tool

    GitHub URL in, running app out. It writes the Dockerfile and the CI pipeline, deploys to Kubernetes, and proposes a fix when the build fails.

    • 7 language templatesSource: docs/paper-claim-verification.md
    • 3 LLM providers in fallbackSource: Groq, then OpenRouter, then Gemini; repo README
    • IEEE paper under reviewSource: Co-author; submission not yet decided

    FastAPI · React · Docker · GitHub Actions · Kubernetes · MongoDB

  3. 03
    Shipped

    ReportGen: a project report from a short form

    Fill in a form, get a complete university-format .docx: cover, certificate, declaration, contents and written chapters.

    • 4.7 s per sample reportSource: Timed on the EC2 server, 2 November 2026
    • 25 of 30 commitsSource: Commit history of the reportauto repository at the October 2026 check

    React · Flask · python-docx · Gemini API · Clerk · AWS EC2

  4. 05
    Working, not yet public

    GitHub compliance evidence collector

    Reads a repository's settings through the GitHub API and maps each check to a SOC 2 criterion, with the raw response saved as proof.

    • 10 checksSource: Checks table in the project README
    • 3 outcomesSource: PASS, FAIL, UNKNOWN; engine.py
    • 8 of 8 tests passingSource: python -m pytest, run on 10 November 2026; tests replay recorded API responses

    Python · httpx · GitHub REST API · pytest

02 · Lab

What I am building now.

Unfinished work, labelled as unfinished. No result appears here until it has been measured.

In progressPyTorch · Open3D · NumPy

PointNet on ModelNet10

A from-scratch PyTorch PointNet that classifies 3D furniture from point clouds. The pipeline is tested end to end. The full training run has not happened yet, so there is no accuracy to show.

  • 13 unit tests passingSource: dev/test_units.py in the repository
  • 811,914 parameters in the baselineSource: Printed by the notebook's model cell
Code
MeasuredNumPy · Tabular Q-learning

Q-learning irrigation agent

Can an agent learn when to water better than the robot's fixed rule? It keeps soil in range far more often than that rule. A hand-tuned threshold still beats it slightly, and I report both.

  • 81.5% of steps in range, against 61.3% for the robot's ruleSource: Results table in the project README: 500 unseen seasons, same seeds for every policy
  • A tuned threshold reaches 85.6%Source: Same results table; the agent does not beat every rule
Built, not field-testedC++17 · CMake

KRISHIBODH optimization engine

C++17 modules for the robot: route planning between plants, a watering scheduler, sensor filters and a non-blocking firmware state machine. Benchmarks use synthetic data and it has not run on the robot.

  • 44 of 44 tests passingSource: Test suite in Krishibodh-Optimization-Edition, built with GCC and CMake
Research onlyEdge LLMs · CAN · COVESA VSS

VehicleGPT

Design research for an offline assistant that answers questions about a vehicle's own data, inside 2 GB of RAM. A dossier of evidence and architecture. No implementation yet.

  • Under 2 seconds, about 2 GB RAM, no cloudSource: Constraints from the project brief, listed in the repository README
Dossier
Private repoPython · scikit-learn · OR-Tools

CoSafar ride pooling

Groups commuters by location and plans shared routes for them. Built for Project Expo 2026.

  • DBSCAN clustering plus OR-Tools vehicle routingSource: optimization.py in the Project Expo backend
Shippedn8n · Graph API · Telegram

Instagram auto-publisher

Send a message to a Telegram bot and it publishes to Instagram. Built as an n8n workflow.

  • n8n, Instagram Graph API v22, Telegram triggerSource: Workflow described in the insta-automation case study
ShippedFlask · Gemini

Notion to Blogger to LinkedIn

Write once in Notion; a Flask service rewrites the post with Gemini and publishes it to a blog and to LinkedIn.

  • Flask service with Gemini rewritingSource: render_deploy/automation.py
PrototypePython · Random Forest · Leaflet

crop-ai

Crop water-stress mapping from Sentinel-1 and Sentinel-2 style imagery, with a map dashboard. Built for the Bharatiya Antariksh Hackathon 2026. It runs on simulated data.

  • Hackathon prototype on simulated dataSource: crop_stress_advisory/README.md; no accuracy is claimed for it
Code

03 · The other half

Engineer by day.
Editor by night.

The film behind this text has no footage in it. Every frame is drawn by code. Short films, design and photographs live in the creator wing.

Enter the creator wing

04 · About

The short version.

I am a final-year computer science student at MVJ College of Engineering in Bengaluru, and a Vahani Scholar. I am also a self-taught video editor, and I keep doing both.

I check my own claims. My résumé has a file that maps every line to the code or page that proves it, and this site does the same thing in public: the Receipts switch in the top bar shows the source of each number.

  1. 2023 to 2027

    B.E. Computer Science, MVJ College of Engineering

    CGPA 9.01 / 10Source: Current CGPA as on the October 2026 résumé

  2. Aug 2026 to now

    Marketing Intern, Aerobiotix India

    Weekly LinkedIn and Instagram content for a hospital air-disinfection companySource: Role as described on the résumé

  3. Leadership

    Team Lead, Vahani AI and Automation Bootcamp

    About 100 scholars, rated 4.64 / 5Source: Participant feedback shown in the bootcamp case study; I co-led it open

  4. Mar to Jun 2025

    Web Development Intern, CEQUE

    Responsive WordPress pages, plugin setup, layout and mobile fixesSource: Role as described on the résumé

  5. Aug 2024 to Feb 2025

    Social Media Intern, Vahani Scholarship Trust

    SEO and hashtag strategy across Instagram and LinkedInSource: Role as described on the résumé

  6. Campus

    GDG On Campus tech team, Design Lead at NIC

    Selected as 1 of 3 from 70+ applicantsSource: GDG On Campus selection, as on the résumé