AI researcher · Quant enthusiast · Builder
Portfolio / Puzzle Lab / LinkedIn / Email / WCA
SASTRA Deemed University · B.Tech CSE (AI & DS), 2024–2028 · Hosur, India
How do you turn noisy observations into better decisions?
That question connects most of my work: exchange-rate forecasting, reinforcement-learning agents for traffic signals, and machine learning for network security. I'm Mithilesh, a Computer Science undergraduate at SASTRA Deemed University, exploring the intersection of deep learning, reinforcement learning, and quantitative finance.
I also build the systems around the models: data pipelines, APIs, dashboards, and applications that make an idea usable beyond a notebook.
| Research lens | Engineering lens | Mathematical lens |
|---|---|---|
| Learn from sequential data and feedback | Connect models to useful applications | Understand uncertainty, signals, and risk |
| DNN–RL forecasting · APT detection | Multi-agent traffic control · Voice-first AI | Alpha research · Probability · Time series |
| Project | What makes it interesting |
|---|---|
| Adaptive Traffic Management | SIH 2025 idea developed into a multi-agent platform: DQN signal control, emergency preemption, routing, MQTT telemetry, and a dashboard. A hard safety layer constrains learned policies. Simulation and hardware integration points are documented separately. |
| CraftHaat | Voice + photo → AI-generated product catalog. Offline capture with a self-hosted speech/image/LLM backend and ONDC payload preparation. A prototype: live ONDC publishing and production authentication remain open work. |
| Exchange Rate OpenEnv | An environment where agents choose whether to accept, replace, or drop currency-feed ticks under missing data, price spikes, and latency. Data remediation, distinct from my forecasting research. |
| Deep Learning Lab | A collection of deep learning models built while working through the fundamentals. |
More from the workbench
- Amazon ML Challenge — challenge work and notebooks.
- Data Science Projects — a collection of my projects.
- PyMC / GSoC 2026 Preparation — preparation repository; not a claim of program selection.
- Prep — coding, DSA, competitive programming, and quantitative-finance preparation.
- Portfolio source — the website behind my work.
My GitHub also includes forks for learning and exploration; those are not presented here as original projects.
DNN–RL exchange-rate forecasting · March 2026–present
Exploring hybrid deep neural networks and PPO policy-gradient reinforcement learning for financial time-series forecasting, incorporating crude oil, gold futures, and NIFTY 50 as macroeconomic signals.
Deep learning · PPO · Multivariate time series · Quantitative finance
Publication status: submitted to IEEE and under peer review; this is not a published-paper claim.
Advanced Persistent Threat detection · February 2026–present
Investigating machine learning and anomaly detection for identifying persistent, sophisticated threats in network activity.
Anomaly detection · Network security · Machine learning
| Area | Technologies & concepts |
|---|---|
| Learning & decision-making | PPO · DQN · Multi-agent RL · Policy gradients · Transfer learning · TensorFlow · PyTorch |
| Language & generative AI | Transformers · BERT / GPT · Hugging Face · RAG · Prompt engineering · Whisper · Ollama |
| Vision | OpenCV · YOLO · Object detection · Image segmentation |
| Data & mathematics | NumPy · Pandas · Matplotlib · Seaborn · Probability · Statistical modeling |
| Quantitative finance | Alpha research · Backtesting · Algorithmic trading · Risk analysis · Portfolio optimization |
| Applications & infrastructure | FastAPI · Django · Flask · React · Flutter · REST / WebSocket APIs · MongoDB · PostgreSQL · Redis · MQTT · Docker |
| Languages & environment | Python · C++ · C · Java · JavaScript · SQL · Git · Linux · Jupyter · Raspberry Pi |
WorldQuant · BRAIN Research Consultant
April 2026–present
Developing and backtesting quantitative alphas on the BRAIN platform using probability and statistical modeling; reached Gold Genius level.
XYlofy AI · AI & Data Science Intern
June–July 2026
Built machine-learning architectures and Python data pipelines, integrating REST APIs and model backends for automation projects.
Freelance AI/ML Developer
2026–present
Building custom AI/ML solutions and full-stack applications, from data pipelines and training workflows to APIs.
SASTRA Deemed University · Student Researcher
February 2026–present
Working on exchange-rate forecasting and APT detection research.
| Milestone | Detail |
|---|---|
| CMI STEMS 2025 | Top 30 in India · Chennai Mathematical Institute |
| WorldQuant BRAIN | Gold Genius level · quantitative alpha research |
| Smart India Hackathon 2025 | RL-based intelligent traffic management · Odisha |
| Mathematics Olympiad | IOQM qualifier · Merit certificate |
| Yale / Coursera | Financial Markets certification · Completed |
| NPTEL SWAYAM | Deep Learning coursework · 2026 |
| Harvard CS50W | Web Programming with Python and JavaScript · In progress |
Generated from my public GitHub contribution calendar. This is a dated snapshot, not a live feed or performance metric. Click either chart for the latest GitHub activity. Refresh instructions are in SETUP.md.
Speedcubing is a hands-on way to think about state spaces, patterns, and legal moves. My WCA profile is 2022ADHI01. Official personal-best singles: 37.70s on 3×3, 13.17s on 2×2, and 6.61s on Pyraminx, from Cubing Returns Bengaluru 2022.
The Puzzle Lab explores playable 2×2–5×5 cube permutations alongside kinetic Pyraminx and Clock geometry.
Open to research collaborations, internships, and freelance AI/ML work—especially where learning systems meet financial data or real-world decisions.
Portfolio · LinkedIn · GitHub · Stack Exchange · Freelancer
Mithilesh Adhinarayanan · State. Move. Result.