I'm

Anushka.

I didn't wait for a PhD to do real research.

I study what happens when intelligence runs without goals — and I build systems that run without internet. From constraint-driven evolutionary frameworks that sustain novelty for 10,000 generations, to offline edge AI navigating at 20+ FPS on a Snapdragon chip. Theory. Code. Silicon. All three. At the same time. Currently seeking Summer 2026 research opportunities.

View Resume
Anushka Sharma profile image

Who Am I?

Anushka Sharma profile picture

I live at the edge of two questions most people treat as separate.

How do you build AI that keeps discovering new things — without being told what to find? And: How do you make it run on a chip that fits in your pocket?

I’m working on both. I’m 21, undergraduate, and I have a first-authored paper under review at GECCO 2026 to show for it.

My research is on open-ended evolution — systems that generate genuine novelty without reward hacking, without explicit fitness functions, without anyone defining what “good” means. My engineering is on edge AI — real-time computer vision on Raspberry Pi, Jetson Nano, Snapdragon NPUs. No cloud. No compromise.

I think the most dangerous AI isn’t the one that fails. It’s the one that succeeds at the wrong thing. My work addresses both halves of that problem.

What I build with:
  • Python · C++ · C · JavaScript · Verilog
  • TensorFlow / TF Lite · PyTorch
  • OpenCV · scikit-learn · XGBoost · SHAP
  • ROS · Raspberry Pi · Jetson Nano · ESP32
  • Snapdragon NPU · Pixhawk PX4 · Arduino
  • Docker · Flask · Streamlit · Git/Linux
  • Evolutionary Computation · Edge AI
  • SLAM · Autonomous Navigation · Sensor Fusion

Where I Have Worked

ML Engineer Intern - SecureVision.Tech
June - September 2024

Real-world computer vision is brutal — low light, motion blur, faces at impossible angles. I made a production system handle all of it.

  • +15% accuracy under adverse real-world conditions
  • 20% lower latency via model quantization and pipeline optimization
  • Stack: Python 3.8, TensorFlow 2.x, OpenCV 4.x
Head of AI Development - ADM Foundation
June - October 2024

Led an entire AI department at a national NGO. Fifteen engineers. Zero missed deadlines.

  • 15-member technical team managed end-to-end
  • 95% on-time delivery rate across all active projects
  • 25% faster turnaround by redesigning the agile workflow from scratch
C++ Developer Intern - ApnaGuide
June - July 2024

Went in, found the bugs, fixed what was slowing things down, and left the codebase measurably better.

  • Critical bugs resolved across 10+ production modules
  • High-severity issues reduced by 30%
  • Runtime efficiency improved by 12% through targeted algorithmic rewrites
  • Stack: C++14/17, STL, GDB, Valgrind, Git
HR Intern - Marpu Foundation
June - July 2024

Streamlined people operations and rebuilt the candidate experience end-to-end.

  • Onboarding time cut from 5 days to 2 days — 60% reduction
  • Managed a pipeline of 100+ candidates with structured workflows
  • Skills: Stakeholder communication, Process design, People management

Education

2023 - 2027
B.Tech in Electronics and Communication Engineering
Banasthali Vidyapith, Rajasthan, India
GPA: 8.89 out of 10.00 (3.56 out of 4.00 equivalent)

ECE is a rare combination: signal processing, embedded systems, and machine learning under one roof. Studying the hardware and the intelligence together — that’s deliberate.

Coursework: Machine Learning, DSA, Probability & Statistics, Linear Algebra, Optimization, Signal Processing, Embedded Systems, Computer Architecture

What’s happened while still in undergrad:

  • GECCO 2026 paper under review — first-authored, two novel metrics, large effect size
  • Top 10 at Qualcomm Edge AI Hackathon from 650+ teams nationwide
  • Round 3 at Google Girl Hackathon 2025
  • Round 2 at IIT Bombay E-Yantra Robotics Competition
  • Published author at 17, international release
Genesis: Emergence — Core Research Repo
2022
Class XII, CBSE
Dewan Public School, Meerut, Uttar Pradesh
GPA: 90% — Physics, Chemistry, Mathematics, Computer Science
The year I finished school was also the year I published my first book. Artificial Intelligence: What It Means to Us (Aegis India, 2023) — written at 17, released internationally. It was not an assignment. It was a statement.

What I Have Built

Genesis Engine — Evolution Without Goals
Python Evolutionary Computation AGI Safety NumPy SciPy
Genesis Engine — Evolution Without Goals
Can a system keep generating genuinely new things forever — without being told what "good" means? I spent quite some time building one that does. 10,000 generations. No fitness function. 58.3% sustained evolutionary activity. The system doesn't optimize. It doesn't maximize. It doesn't know what success looks like. And yet — it keeps creating novelty. This is a first-authored GECCO 2026 submission. Two new metrics: CARP and PNCT — introduced because the field didn't have the vocabulary for what I was measuring.
Research Repo
VisionIRL — AI Navigation, Zero Internet
TensorFlow Lite YOLOv5 INT8 Snapdragon NPU OpenCV Edge AI
VisionIRL — AI Navigation, Zero Internet
A dead zone shouldn't mean a blind person is stranded. VisionIRL doesn't care about your network signal. Fully offline. Zero cloud dependency. Real-time object detection at 20+ FPS with under 50ms end-to-end latency — on a mobile chip. YOLOv5s INT8 on Snapdragon NPU. TensorFlow Lite pipeline. Built it, deployed it, took it to Qualcomm's national Edge AI Hackathon. Top 10 from 650+ teams. Youngest team. Most appreciated.
Hackathon Repo
SynthPredict — 98% Accurate, Instant
XGBoost scikit-learn SHAP MLP RTL Design
SynthPredict — 98% Accurate, Instant
RTL synthesis takes hours. SynthPredict does it in milliseconds. 98% prediction accuracy for combinational depth estimation using an ensemble of MLP and XGBoost — with full SHAP explainability so engineers understand *why*, not just *what*. Engineers get hours back. Every design iteration. Google Girl Hackathon 2025 — Round 3 nationally.
View Project
ApnoBot — The Bot That Reads the Room
Python Flask OpenAI API Telegram Bot API Prompt Engineering
ApnoBot — The Bot That Reads the Room
Not just a chatbot — a conversation partner that adapts. AinoBot detects your mood, switches personas, remembers context, and responds like it actually understands what you're going through. GPT-3.5-turbo with a sliding context window and real-time sentiment analysis. Conversation that moves with you, not past you.
AutoPage — Merged Open Source PR
Python NLP Open Source HTML/CSS
AutoPage — Merged Open Source PR
Shipped a feature to production. Used by 1,000+ developers worldwide. Still running. Built a domain detection system for AutoPage that automatically adapts documentation style for API manuals, research papers, and product specs — based on content type alone. PR merged into AutoLab-SAI-SJTU. Not a proposal. Production code.
AutoPage Repo

Proof of Work

GECCO 2026 — First-Authored Paper Under Review
Sustained Evolutionary Activity Without Fitness Functions. 10,000 generations. 58.3% sustained activity. Cohen's d = 1.47. p < 0.01 across 12 independent runs. Two new metrics introduced — because the field didn't have the vocabulary to measure what I was seeing. One big idea, rigorously tested.
Qualcomm Edge AI Hackathon 2025 — Top 10 Finalist
650+ teams. One shot. We built a fully offline AI navigation system for the visually impaired — no cloud, no internet, no compromise. Youngest team in the competition. They called us "most appreciated." The chip didn't know it was supposed to be limited. Neither did we.
Google Girl Hackathon 2025 — Round 3 Nationally
Advanced to Round 3 with SynthPredict — an ML model predicting RTL combinational depth at 98% accuracy. RTL synthesis takes hours. SynthPredict takes milliseconds. Engineers get that time back. Every design iteration.
IIT Bombay E-Yantra Robotics — Round 2
Led a 4-member team building a live autonomous drone mapping system for disaster management. SLAM, waypoint navigation, obstacle avoidance — all working together in real time. When it matters most is exactly when it has to work.
Published Author at Age 17
*Artificial Intelligence: What It Means to Us* — published internationally by Aegis India in 2023. Written because I believed AI was too important to stay inside academic papers. Ethics, consciousness, societal stakes — for everyone, not just researchers. Written at 17. Not as a school project. As a statement.
First Prize — Speech, Poetry & Ad-Mad
First Prize in scriptwriting, speech, radio jockeying, and self-written poetry. Building something matters. Making the world care about it matters just as much. I've trained both skills — deliberately.
Winner — ISRO Essay Writing Competition
Space, AI, and systems thinking all live in the same universe. Literally.
YouTube Science Communicator
Good research deserves to be understood — not just cited. Technical content on AI, evolutionary computation, and embedded systems for an audience that wants depth, not just headlines.

Let's Talk

I respond to emails worth responding to.

If you’re researching open-ended evolution, AGI safety, or edge intelligence — I want to talk. If you think constraint-driven systems are the wrong approach to novelty — I especially want to talk. If you want to work with someone who’s already solving the problems you’re working on — you know where to find me.

Write to me