My 2027 plan: weekly tasks, contests, problem bank, recall and rating tracker. Password required.
A 20-year-old engineer-in-training building across Electrical Engineering, Machine Learning Systems, and the odd competitive-programming contest in between — currently researching differential privacy and Byzantine-robust distributed estimation, with side quests in quant finance, analytics and robotics.
I'm Barad Dhanush — a tech enthusiast who ended up studying Electrical Engineering and never quite stopped exploring everything adjacent to it. My core interests sit across Competitive Programming, Math, Finance, Machine Learning and AI, and the electronics/electrical core that got me here in the first place — and over the past year it has pulled me into Quant (a regime-detection model and a portfolio competition), Consulting-style analytics (a pricing bootcamp), and a student-run mentorship product at Enactus.
Right now that curiosity has a fairly specific shape: making distributed estimation and Machine Learning Systems stay accurate and private even when some of the nodes in the network are faulty or adversarial.
July 2025 — July 2030 (expected) · Currently a sophomore
A summer spent finding and fixing a gap in a published privacy-preserving estimation scheme: what happens when a relay is not just unreliable, but malicious.
Building on PRICER (IEEE Trans. Signal Process., 2024), a differentially private, semi-decentralized mean-estimation scheme. I showed that a single Byzantine relay can make its linear aggregation error unbounded, then designed BRICER to close that gap while keeping the original privacy guarantees.
Actively looking right now — happy to talk to labs or industry research teams.
Prof. Chiang has invited me back to continue the collaboration at NCCU next summer.
Analytics, quant, game-AI and robotics projects from bootcamps, competitions, hackathons and coursework. Walkthroughs are available on request.
A game-playing agent for Settlers of Catan. Held 1st place on the public leaderboard for a long stretch of the contest, with a score of 0.836 vs F and 0.613 vs O (mean 0.725) at 249 CPU ms per game.
Finished in the top 15 of 64 teams as the youngest team, against 3rd- and 4th-year seniors.
A pricing engine for a retailer with 49 SKUs in 8 categories, built from five datasets merged into 53,704 cleaned transactions. Estimated SKU-level price elasticity with log-log OLS (45 of 49 significant), then optimized prices by grid search under cost-floor, competitor-ceiling and volume-drop guardrails.
Simulated +8.30% gross margin (≈ Rs. 2.6 Cr annualized) for only −2.93% volume. Delivered an Excel dashboard and executive deck.
A Gaussian Hidden Markov Model that detects four market regimes (Crisis, Bear, Bull, High Growth) from Indian equity, gold and bond ETFs and switches portfolio allocation accordingly. The 4-state model was chosen by BIC, and validated with expanding walk-forward backtests and transaction costs on every rebalance.
Best risk-adjusted results among all benchmarks, with about half the volatility and max drawdown of equal-weight and static 60/40.
A Rs. 1.5 Cr, 10-year portfolio for a family with four goals: liquidity, property, retirement and growth. Built six goal-linked baskets, and screened the 750-stock Nifty Total Market on ROIC, D/E, revenue CAGR and free cash flow to pick 11 equities.
Stress-tested under an RBI rate-cut cycle and a defence-exports boom, including tax impact on net returns.
Punjab's detected farm fires fell ~70% from 2023 to 2024 while burned area stayed flat, so the real target is deadline-aware, non-burning residue management. Designed Saanjhi Vaadi, a village-cluster layer that pools farmer demand into date-specific machine-hours and exposes shortages.
Owned the time-windowed fleet scheduling model (mixed-integer program, OR-Tools CP-SAT), benchmarked against manual, first-come-first-served and earliest-deadline-first dispatch, with Punjabi speech intake and a six-village pilot design.
A wired four-wheeled robot that grips a cylindrical load and lifts it along the Z-axis like a miniature forklift. Acrylic chassis designed in Fusion 360 and laser cut, with a 3D-printed rack-and-pinion gripper.
An Arduino Nano reads a PS2 joystick and sends it over a wired UART link (RJ11) to an ESP32 driving 5 DC motors and a servo through two L298N drivers. The transmitter is soldered on a PCB.
A wireless-controlled robot on a Raspberry Pi Pico W, ESP-01 and MPU6050 IMU. Built a custom soldered board with a 12V LiPo, voltage regulators and an L298N driver for four 200 RPM DC motors, for stable torque and smooth maneuvering.
Contributing to an AI-guided mentorship platform (Next.js, Supabase, Claude API) that matches students with alumni mentors for high-stakes academic and career decisions. Working on schema design, decision taxonomy and embedding-based mentor matching.
8-week programme covering segment trees, Fenwick trees, DSU, binary lifting, bitmask and tree DP, max-flow, bipartite matching, string algorithms and computational geometry. Practiced in timed Codeforces and AtCoder contests and built a personal template library.
Completed the introductory course on semiconductor fabrication.
Where most of my thinking time goes.
What I do when I close the laptop.
Open to remote or corporate research internships now, and returning to NCCU for research in Summer 2027. Reach out any time.