ayush = {
"name" : "Ayush Kumar",
"degree" : "B.Tech CSE (AI & ML) @ KIIT University",
"year" : "2nd Year | Class of 2028",
"cgpa" : 8.4,
"focus" : ["Applied ML", "Feature Engineering", "Model Explainability",
"AI-powered Web Apps", "Data Products"],
"currently" : "Building production-grade ML pipelines & hunting for internships",
"fun_fact" : "My best model predicts F1 podiums with 98.61% top-3 accuracy 🏎️"
}16 merged PRs across 4 repositories · GSSoC 2026 · May–Aug 2026
Twelve of these landed in one codebase — AegisGraph Sentinel, a GNN-based
fraud detection platform — where I owned the explainability and model-trust layer.
Several PRs replaced explanation and monitoring components that were returning
fabricated values: mock counterfactuals scored with random.uniform, drift
baselines drawn from unseeded np.random, and a data loader that advertised
leakage prevention while splitting randomly.
| Repo | Merged | Area |
|---|---|---|
| AegisGraph Sentinel 2.0 | 12 | XAI pipeline, calibration, training integrity |
| AegisAI | 2 | API endpoints, integration tests |
| CodeGraphContext | 1 | CLI error propagation |
| HELPDESK.AI | 1 | Unit tests for classifier service |
7 of 16 rated Advanced by maintainers · 3 rated quality:exceptional or clean
→ Case study: 12 PRs into AegisGraph Sentinel · → All merged PRs
🏎️ RaceIQ — F1 Race Intelligence Platform
End-to-end Formula 1 race prediction platform — the kind of project that makes ML interviewers lean forward.
| Metric | Value |
|---|---|
| 📦 Dataset | 2,979 driver-race records · 149 races · 2018–2024 |
| 🎯 Finishing Position MAE | 0.4447 (XGBoost) |
| 🏆 Top-3 Prediction Accuracy | 98.61% on 2024 test set |
| ⚙️ Stack | FastF1 · XGBoost · SHAP · Optuna · Streamlit |
What's under the hood:
- Multi-target models — finishing order, podium probability, points probability, DNF risk
- SHAP explainability layer for per-race feature attribution
- Temporal train/test split (no data leakage)
- Full experiment tracking and Streamlit dashboard
🌍 EcoSentinels — AI Environmental Intelligence Platform
AI-powered ecosystem risk monitoring platform built for the Kaggle Gemma 4 Good Hackathon — real-time AQI data, RAG-grounded field briefings, and multi-language community alerts.
| Metric | Value |
|---|---|
| 🤖 LLM | Gemma 4 (26B) via Google AI |
| 🧠 RAG | ChromaDB · 78 chunks · 7 knowledge docs |
| 🌐 Agents | DataAgent · AnalystAgent · AlertAgent |
| ⚙️ Stack | FastAPI · React · TypeScript · LangChain · Vercel |
What's under the hood:
- Multi-agent pipeline — live AQI + weather ingestion, RAG retrieval, alert generation
- Community field briefings with threat level, evidence, and immediate actions
- Multi-language translation into 8 languages via Gemma 4
- Offline fallback — frontend degrades gracefully when backend is unreachable
🛡️ TrainSentry — AI-Assisted ML Training Log Analyzer
Developer tool that automates the first debugging pass of ML training runs — rules engine for anomaly detection, Gemini-powered debugging narratives, CLI and Streamlit dashboard.
| Metric | Value |
|---|---|
| 🔍 Detects | Overfitting · Val spikes · Stagnation · Metric drift · Leakage signals |
| 🤖 AI Reports | Gemini 1.5 Flash — root cause analysis + prioritised action plan |
| 🧪 Testing | pytest suite + GitHub Actions CI |
| ⚙️ Stack | Python · Pandas · Streamlit · Plotly · Gemini API |
What's under the hood:
- Rules engine detects 5 training failure modes with evidence and recommendations
- AI report mode sends structured findings to Gemini for cross-finding root cause analysis
- Graceful fallback — works fully offline without an API key
- CLI with
analyzeandcomparesubcommands +--saveflag for report export
Languages
ML / Data Science
AI / LLM
Visualization & Backend
Tools
| Paper | Venue | Status |
|---|---|---|
| TensorFlow Quantum For Strategic Simulations : Integrating Quantum Game Theory and AI | ICDCIT 2026 — International Conference on Distributed Computing and Internet Technology | ✅ Accepted · Proceedings copy incoming |
- 📚 2nd Year B.Tech CSE (AI & ML) @ KIIT University — CGPA: 8.4
- 📝 Published at ICDCIT 2026 — proceedings copy en route
- 🏗️ Building production ML pipelines with experiment tracking + full test suites
- 🎯 Actively seeking ML / AI Engineering internships (Summer/Fall 2026)
Open to collaborations, research projects, and internship opportunities in ML/AI.