I'm Chandradithya — I build systems that run in production and models that run in the real world. Currently making agentic AI at Emerson and advancing defence imaging research under a DRDO grant. CS final year, VNRVJIET.
Obsessed with hard problems.
I got into software because I wanted to understand how things actually work — not just use them. That curiosity dragged me through compilers, through kernel internals, through graph algorithms at 2am before a contest. It still does.
These days I split time between building agentic AI at Emerson — where I've shipped invoicing automation and a session-isolated backend serving thousands of concurrent users — and research under a DRDO-IRDE grant, where I'm pushing the limits of how well a neural network can see through fog, rain, and haze in real-time defence optics.
Before all that, I spent most of 2024 as a core contributor to Sarvadrushti, an open-source vision project that went from a side experiment to an institutionally funded research project. I've merged PRs into multiple repositories. Small contributions, but I learned more from reading those codebases than from most coursework.
I'm looking for the kind of role where the technical bar is high and the problems are genuinely unsolved. CGPA 9.05. Strong on systems, strong on ML, strongest when both are required at once.
Tools I think in.
From systems programming to orchestration to inference — the full picture.
C
C++
Python
Java
JavaScript
Rust
Bash
Spring Boot
React
Svelte
Express
FastAPI
Kafka
PyTorch
AWS
Docker
Kubernetes
Grafana
Terraform
GraphQL
LangChain
MySQL
MariaDB
MongoDB
Redis
PostgreSQL
Where I've worked.
- Shipped a full-stack invoicing system that cut manual billing effort by 90% — React frontend, FastAPI backend, C++ computation engines for high-throughput processing.
- Architected a session-isolated agentic backend that handles 1,000+ concurrent users at TTFT p95 under 3 seconds, running entirely on local infrastructure for air-gapped security compliance.
- Built a RAG pipeline (LangChain + FAISS) with custom chunking and retrieval strategies — meaningfully improved accuracy on internal knowledge queries.
- Developing Transformer-based models that restore weather-degraded optical sensor images with 92% structural similarity — practically useful in adverse field conditions.
- Cut inference latency by 90% through network pruning and aggressive pipeline optimisation, making real-time deployment viable.
- 150%+ gain in scene context analysis accuracy over baseline using custom image enhancement algorithms.
- Built and benchmarked Vision Enhancement algorithms for paramilitary use — fog, rain, and haze removal from live optical feeds.
- REST APIs and ingestion pipelines at sub-500ms latency with integrity checks and automatic fallback handling.
- Modified CycleGAN architecture improved model robustness by ~40% in low-visibility scenarios. Directly contributed to the grant proposal that became DFTM/034.
Things I've shipped.
Production constraints, real stakes, non-trivial problems.
K8s
React
Docker
Redis
Rust
Redis
Svelte
Python
Bash
Python
YOLO
OpenCV
Streamlit
Python
LSTM
Pandas
Pytorch
FastAPI
React
llama.cpp
Python
Let's build something.
Open to SDE, AI/ML, systems, and fintech roles. Also happy to just talk about a hard problem.