I build systems that take data in, run heavy computation, and put the result in front of a user: a model or solver in the middle, a pipeline feeding it, an API and dashboard on top.
Final-year B.Tech IT student at SRMIST Chennai, currently an AI Intern at Applied Systems Research & Technology, leading a small team of undergrads on an AI infrastructure observability platform. I'm especially drawn to the GPU / high-performance computing side: simulation, inference speed, and squeezing more out of hardware.
| 🎓 Education | B.Tech Information Technology, SRMIST Chennai (2023–2027) |
| 💼 Now | AI Intern, Applied Systems Research & Technology Pvt Ltd |
| 🎯 Open to | AI Engineer · Data Engineer · Forward Deployed Engineer · SDE · GPU/HPC-adjacent roles |
| 📍 Base | Chennai, India |
| Role | Why | Proof |
|---|---|---|
| AI Engineer | Train, serve, and guard models in real systems | YOLOv8 multi-camera pipeline · churn model (ROC-AUC 85.03%) · LLM guardrails in REVORA |
| Data Engineer | Ingest, model, and serve data reliably | Relational ETL → MongoDB → star schema · TimescaleDB metrics in AISO |
| Forward Deployed Engineer | Take a messy real-world problem to a working deployed solution | Disaster-response platform (SIH 2026) · industrial PPE monitoring · payments recovery app |
| SDE / Backend | Clean APIs, auth, queues, containers, CI | FastAPI + Redis + WebSockets + JWT · Docker · GitHub Actions |
| GPU / HPC | Simulation and inference are where I want to go deeper | Hydrodynamic + SPH solver pipeline · real-time CV inference · see Roadmap |
AISO: AI Infrastructure Observability Platform
- Built on FastAPI · PostgreSQL · TimescaleDB · Redis
- Lead a small team of undergraduate contributors on the codebase
🌊 JalRakshak-HD: Dam-Break & Disaster-Response Intelligence Platform
Smart India Hackathon 2026 · Problem Statement SIH26161 · Team of 6 My role:
A configuration-driven pipeline from dam geometry to breach hydrograph, 2D flood simulation, near-field particle (SPH) simulation, hazard classification, and humanitarian exposure screening, served through a React GIS dashboard. New dams are onboarded with YAML, not code changes.
| 2D flood model | D-Flow FM · 818 km² domain · 30 h simulated · 181 output frames |
| Near-field model | DualSPHysics · ~11K particles · first 1.5 km below the breach |
| Exposure screening | Population, buildings, roads, bridges, critical facilities, 6 response sectors |
| Engineering | FastAPI + React 19 + TypeScript · 57 tests · mass-balance residual 0.0052% |
| Honesty by design | Anything not computed is shown as NOT_RUN, never filled in |
FastAPI React 19 TypeScript D-Flow FM DualSPHysics GeoPandas Google Earth Engine Sentinel-1
🔭 Next on its roadmap: on-demand solver runs on an HPC queue and full 3D SPH on GPU.
👷 Worker Safety Detection v2: Real-time PPE Compliance
Multi-camera YOLOv8 pipeline: threaded capture per camera → Redis pub/sub → severity classification → WebSocket alerts to a React dashboard, with JWT + OTP auth, snapshots, email alerts, and one-command Docker deployment.
| Pipeline | Per-camera threads · spatial helmet/vest matching per worker · alert cooldown throttling |
| Stack | YOLOv8 · OpenCV · FastAPI · Redis · SQLAlchemy · React 19 + TypeScript · Docker |
| Metrics |
📉 Telco Customer Churn: Prediction + Executive Dashboard
SQL analytics → EDA → feature engineering → Random Forest → 7-page Power BI dashboard with a retention what-if simulator, on 7,043 customers.
ROC-AUC 85.03% · Accuracy 80.7% · built with a team of 3
SQL Pandas Scikit-learn Power BI DAX
🗄️ Enterprise Data Engineering Pipeline: Multi-Database Architecture
A retail scenario across three storage layers: relational ETL, MongoDB product catalog, and a star-schema warehouse (fact + dimension tables) for analytical queries.
| Scale | |
| Stack | Python · SQL · PostgreSQL · MongoDB |
🚗 FuelIQ: Vehicle Intelligence Platform (in development)
Flutter app + FastAPI backend: multi-vehicle garage, fuel-efficiency analytics, service reminders, and on-device OCR receipt scanning. Backed by PostgreSQL 16 (partitioned tables, materialized views), Redis, and MinIO, and documented end to end: PRD → architecture → DB design → scale plan.
Flutter FastAPI PostgreSQL Redis MinIO Docker GitHub Actions
- REVORA: AI revenue recovery for failed payments (Razorpay Buildathon 2026). LLM proposes actions; a deterministic policy gateway approves them. FastAPI · Next.js · Gemini · live on Vercel/Render
- AI Contract Analyzer: multi-agent contract review with LangGraph
- SentinEdge: enterprise fraud risk pipeline with tiered review queues
| Languages |     |
| AI / ML |     |
| Data |       |
| Scientific / Geo |    |
| Backend |    |
| Frontend |    |
| DevOps |    |
What pulls me here: the moment a system stops being "does it work?" and becomes "how fast, and on what hardware?"
Where it already shows up in my work
- Running physics solvers (D-Flow FM, DualSPHysics) as stages of a larger pipeline in JalRakshak-HD
- Real-time inference under frame-rate and latency limits in the PPE detection system
What I'm building toward → see Roadmap below.
- GPU vs CPU inference benchmark for the PPE detector (FPS, latency, batch size), then export to ONNX / TensorRT
- Run the near-field SPH case on GPU and publish timing vs CPU
- CUDA fundamentals: write and profile a few kernels (reductions, matrix ops)
- MPI / multi-process basics on a small simulation workload
- PySpark layer on the data-engineering pipeline
- Azure deployment + DP-900
- Harden FuelIQ (tests, CI, deployment)
Checked items get a link to the repo or write-up.
Hiring for 2027 graduates in AI, data, backend, or GPU/HPC-adjacent roles? Let's talk.
