#1

AI engineer. Building systems that think.

IIT (ISM) Dhanbad, Electrical Engineering. I flew to London, won a global AI safety hackathon, published the paper, and came back with a fellowship. Currently founding Grind IRL and finishing a patent-pending quantum RL system for real-world optimization.


Selected Work

Grind IRL gamified productivity platform — 50+ users, live
Quantum EV Optimizer patent-pending hybrid quantum-RL for EV allocation
LLM Deception Detector global hackathon winner — AI safety research, Apart Research
all projects →

Writing

Why I Take Career Advice From an Anime Villain and a Twitch Streamer
Jun 2025
Finding Deception in Language Models
Dec 2024 ↗
all writing →

Projects

Grind IRL 2025

Full-stack gamified productivity platform. Built, shipped, 50+ active users.

Quantum EV Optimization 2025

Patent-pending hybrid quantum-RL for EV charging allocation. Cuts fleet wait times ~25%.

LLM Deception Detection 2024

Global hackathon-winning AI safety research on detecting sandbagging in GPT-3.5.

Multi-Agent Market Research 2024

Autonomous GenAI use case generator using CrewAI. Industry profiling in minutes, not days.

Yelp Sentiment & BI 2024

Predictive pipeline across 8M reviews surfacing patterns that actually predict business success.

Brainstellar Tracker 2024

Chrome extension for quant interview prep tracking. Published to the Chrome Web Store.

← projects

Grind IRL

Productivity apps are usually to-do lists with branding. Grind IRL turns your actual goals into RPG mechanics — you level up, build stats, clear quests. I built and shipped the full stack, acquired 50+ active users without ads, and I'm still iterating on it live. The app exists because I wanted something like it and nothing else was close.

Built with React, TypeScript, Firebase, and GCP.

← projects

Quantum EV Optimization

Every city is placing EV chargers based on where demand is today. That's wrong — you need to optimize for where fleets will be tomorrow, under load, across the grid. I formulated the allocation as a QUBO problem, used Graph Attention Networks to compress urban grid data into quantum-compatible embeddings, and built a hybrid QAOA + Grover-style pipeline that cuts total fleet wait times by ~25% at a 100% success rate under load. Patent pending. Advised by Prof. Sukanta Halder.

Built with Python, PennyLane, Qiskit, and PyTorch.

← projects

LLM Deception Detection

Language models sandbag — they deliberately underperform on evaluations. Until recently there was no reliable metadata signal for when it was happening. I analyzed 1,200 GPT-3.5 responses and 15,000+ data points, looking at response time and token usage patterns during known-deceptive outputs. Found a consistent 25% reduction in token usage when the model was sandbagging. Won the Apart Research global AI safety hackathon in London with this. Published the paper. Got the fellowship.

Built with Python, Pandas, and HuggingFace.

← projects

Multi-Agent Market Research

Market research reports that take a consulting team three days to produce contain 80% of the same structure. I orchestrated a CrewAI multi-agent workflow — Researcher, Analyst, Collector — using Tavily API to automate industry profiling and dynamically generate actionable GenAI use cases per sector. What takes analysts a week now takes minutes.

Built with Python, CrewAI, Tavily API, and Streamlit.

← projects

Yelp Sentiment & Business Intelligence

Yelp has 8 million reviews and most analysis stops at average star ratings. I built a predictive pipeline using VADER for sentiment at 71% precision, then applied XGBoost and logistic regression to 200,000+ data points to identify statistical patterns that actually predict whether a business thrives or fails. The patterns were counterintuitive — review length and response behavior matter more than raw sentiment score.

Built with Python, XGBoost, Scikit-learn, and Pandas.

← projects

Brainstellar Progress Tracker

Brainstellar is where quant traders practice logic puzzles before interviews. It has no progress tracking — you either remember what you've solved or you don't. I built a Chrome extension that does. Mark problems done, track your history, know where you stand. Published to the Chrome Web Store.

Built with JavaScript and the Chrome Extensions API.

Writing

Why I Take Career Advice From an Anime Villain and a Twitch Streamer Jun 2025
Finding Deception in Language Models Dec 2024 ↗
← writing

Why I Take Career Advice From an Anime Villain and a Twitch Streamer

I don't really vibe with "follow your passion" or "believe in yourself" type advice. It feels generic. It could've been said to anyone. But there are two people — one fictional, one real — whose mindset I actually think about when things get hard.

The first one is Bakugo Katsuki. He's the rival character in My Hero Academia, and in most stories, the rival character is the lesson — the cautionary tale about what arrogance costs you. But Bakugo doesn't lose the plot. He stays at the top. He screams about winning and then goes and wins. And what I like about him isn't the arrogance — it's the clarity. He knows exactly what he wants and he never lets himself forget it, not for one second. I respect that. I want that. That kind of locked-in.

The second one is Stable Ronaldo. If you don't know him — he's a Twitch streamer who was huge, then became basically irrelevant, and instead of pivoting or quitting or reinventing himself as something new, he just kept going. Same thing. Every day. And then the numbers came back. He's massive now. I watch him for the content but I think about him for the mindset. He didn't need the validation to keep working. That's the thing.

Right now I don't have a job. I'm applying. I'm building. Some days nothing moves. But I've got a very specific picture of where I'm going — and I'm not changing it just because it's taking longer than I thought.

I want to make my parents proud. I want to be the kind of engineer my younger brother looks up to. That's not poetic — that's just the actual reason I keep going.

Anyway. Back to building.

Siddharth Reddy Bakkireddy
AI Engineer  ·  IIT (ISM) Dhanbad  ·  besidd.me
besiddharth1@gmail.com  ·  +91 77020 46744  ·  linkedin.com/in/bsiddharth1
Education
IIT (ISM) Dhanbad Expected July 2026
B.Tech, Electrical Engineering  ·  CGPA: 7.55
DSA (C++), OOP, Computer Networks, DBMS, OS, Soft Computing, Probability & Statistics
Experience
GRIND IRL May 2025 – Present
Founder & Lead Developer
  • Engineered and launched grindirl.netlify.app — a gamified full-stack productivity platform.
  • Bootstrapped to 50+ active users through rapid iteration and direct user feedback loops.
Apart Research Jan 2026
Judge — AI Manipulation Hackathon
  • Invited to judge an international technical hackathon focused on AI safety and manipulation vectors.
  • Evaluated technical feasibility and mitigation strategies from global developer teams.
Zetheta Algorithms Private Limited Apr 2025 – Jun 2025
AI Software Engineer Intern  ·  Remote
  • Developed Python modules to parse resume data and engineered LLM prompt templates to dynamically generate technical interview questions mapped to candidate skills.
Apart Research Jun 2024 – Dec 2024
AI Safety Research Fellow / Lab Member  ·  London, UK
  • Secured fellowship after winning the onsite Apart Research hackathon in London.
  • Analyzed 1,200 GPT-3.5 responses (15,000+ data points) to detect deceptive sandbagging behavior.
  • Discovered and documented a 25% reduction in token usage during deceptive outputs.
  • Co-authored and published: "Deception Detection in GPT-3.5 Turbo: A Meta-Data Based Approach."
Projects
Quantum EV Optimization 2025
Patent-pending hybrid quantum-RL for EV charging station allocation. QUBO formulation via Graph Attention Networks; QAOA + Grover pipeline reduces fleet wait times ~25% at 100% success rate under load.
Grind IRL 2025
Full-stack gamified productivity platform turning habits into RPG mechanics. React, TypeScript, Firebase, GCP. 50+ active users.
Multi-Agent Market Research System 2024
CrewAI multi-agent workflow (Researcher, Analyst, Collector) using Tavily API to automate industry profiling and generate GenAI use cases per sector.
Yelp: Sentiment Analysis & Business Intelligence 2024
Predictive pipeline for 8M reviews (71% precision via VADER); XGBoost and Logistic Regression on 200K+ data points to surface business success signals.
Skills
C++, Python, JavaScript, TypeScript, SQL, React, Bash, PyTorch, Qiskit, PennyLane, CrewAI, Firebase, GCP, Git, Scikit-learn, Pandas, NumPy, HuggingFace
Achievements
  • Won 3rd place ($300) in the Global AGI Deception Detection Hackathon by Apart Research (London, 2024)
  • Selected as Judge — Global AI Manipulation Hackathon, Apart Research (Jan 2026)
  • Codeforces Expert (1642)  ·  LeetCode Knight (1817)  ·  Global Rank 675, Codeforces Round 185
  • Patent draft: Graph-Based Optimization for EV Charging Allocation using RL and Quantum Algorithms
  • Published Chrome extension: Brainstellar Progress Tracker
  • JEE Advanced — top 1% of 1,000,000+ engineering applicants in India
Links
besidd.me  ·  github.com/itsbsiddharth  ·  linkedin.com/in/bsiddharth1  ·  x.com/be1sidd