sampling.portrait() · steps 1000 → 0
I build AI agents and the machine-learning systems underneath them. Final year at GGSIPU, Delhi, and always halfway through a book.
01 — about.md
I build AI agents, train models, and write the low-level code that runs them.
attention · hover a word
I've interned at IBM, where I built multi-agent systems; at the Department of Mathematics, IIT Delhi, where I worked on portfolio optimisation; and at ReferralMe, where I automated their referral process.
I grew up in Chittorgarh, Rajasthan, a city of forts, waterfalls and long summer afternoons.
02 — work
tap: int8A ReAct-style agent that breaks a question into sub-queries, searches the web, and writes a cited report.
−70%manual research time (n=20)
hover the screen. it’s the real onetap the screen. it’s the real one
03 — research
I work on predictive maintenance: forecasting when an engine will fail, and using reinforcement learning to decide when to service it. Here’s the paper, with my notes in the margin, and under it my book chapter and thesis.
6th Int. Conf. on Advancement in Electronics & Communication Engineering, 2026 · accepted
Tanu Shree Gupta
University School of Automation & Robotics, GGSIPU, Delhi, India
Abstract—We present a decoupled two-stage deep reinforcement learning framework for prescriptive predictive maintenance of turbofan engines. An attention-enhanced GRU–LSTM model predicts Remaining Useful Life (RUL)1, and a Double DQN maintenance policy2 acts on that prediction through a Markov decision process. For edge deployment, post-training INT8 quantisation cuts the memory footprint by 71.7% and speeds up inference 5.18×3, with only a 1.35% increase in RMSE. Across 200 simulation runs, the agent eliminates catastrophic failures4 while extending average equipment usage to 19.5 cycles.
Index Terms—predictive maintenance, remaining useful life, deep reinforcement learning, TinyML, quantisation.
Code and figures: github.com/gtanushree/IIoT-Predictive-Maintenance
04 — history
2023
augfirst week at GGSIPU. new city, new everything.
2024
decSmart India Hackathon finals. we made it.
2025
julIBM: five agents that talk to each other.
julalso IIT Delhi, maths dept. markets are noisy time series.
octchair of ACM-W.
2026
febGATE: 411.
marcaptain of the basketball team.
a paper accepted and a Springer chapter.
junReferralMe: automated the boring parts.
now
thesis, and looking for what’s next.
— Tanu
05 — outside work
Everything I reach for, filed in the card catalogue. Pull a drawer.