I'm a dual-degree student at IIT Madras and Punjab Engineering College who spends most of his time shipping backend and machine learning systems. Recently that's meant an algorithmic market signal platform serving paying users, a Rust vector index that outperforms FAISS-HNSW on build time and memory, and first-author research on electricity load forecasting. I care about systems that hold up under real load and results you can measure.
Engineered scalable market intelligence using Python OOP models, processing live OHLCV data for 200+ symbols across 8+ algorithmic pipelines.
Designed asynchronous Celery/Redis workflows for sub-2s latency, ensuring reliability via rigorous unit testing and automated AWS deployment.
Implemented Cloudflare security and DDoS protection, delivering a highly available production platform for paying users.
Python
Redis
Celery
Docker
Firebase
AWS
June 2025 — July 2025
Intern · Annam AI (IIT Ropar collaboration)
Led API definition and implementation for a multilingual AI chatbot with context from personalized satellite data reports.
Trained a CNN model for crop disease detection using data aggregated and cleaned from 5 distinct sources.
Automated satellite data report generation (Sentinel-2, Landsat-8) via Google Earth Engine on GCP.
Python
FastAPI
Next.js
GCP
Projects
A couple of the things I've built. There are 15+ more on GitHub.
AIGVS
June 2026 — July 2026
Novel vector index for RAG retrieval
Built a partitioned (non-graph) vector index with hierarchical routing and SIMD-quantized scanning, benchmarked against FAISS on 1M × 384-d MSMARCO embeddings.
Reached 0.9998 Recall@10 (vs. 0.9992 for FAISS-HNSW) with 32× faster index build and 3.1× lower memory than HNSW.
Supports O(1) deletion (1.4 µs) via a lock-free segmented architecture.
Rust
PyO3
SIMD
Voice Emotion & Urgency Detection
May 2025 — June 2025
Multi-speaker audio classification pipeline
Built a hybrid CNN + BiLSTM model classifying multi-speaker emotion across 6 classes, with 500+ background-sound recognition and speaker diarization.
Made an end-to-end audio pipeline — Whisper transcription into 5-level urgency classification — using LLMs via LangChain.