I am a Full Stack Developer building complete, production-grade applications: from interactive client interfaces in React and Three.js to high-throughput distributed backends and low-level storage engines. I work comfortably across the entire stack, whether designing modular REST/SSE APIs, implementing real-time WebSockets, managing relational and document databases, or squeezing throughput out of single-threaded C++ event loops.
- Education: Computer Science Undergraduate (B.E., Class of 2027), MVJ College of Engineering, Bengaluru (CGPA: 8.0/10).
- Industry Experience: Former Software Engineering and Research Intern at the Computational Intelligence (CINT) Lab, Indian Institute of Science (IISc), Bengaluru.
- End-to-End Delivery: Built and deployed full-stack cloud applications on AWS EC2, engineered high-throughput Kafka/PostgreSQL ingestion pipelines, and wrote a C++17 RESP2 in-memory datastore matching 83% of Redis 7.0 throughput.
Indian Institute of Science (IISc), Computational Intelligence (CINT) Lab | Bengaluru, India
- CINT Lab Research Platform (Sole Full-Stack Developer): Designed, built, and deployed the laboratory portal end-to-end using React 18 on the frontend and Node.js, Express, and MongoDB on the backend. Secured authentication with JWT and implemented role-based access control (RBAC) separating administrative workflows from client portals. Deployed on AWS EC2 with continuous integration and deployment on git push (75 registered researchers, 50 active monthly).
- YogaBud Full-Stack Refactor: Transformed a monolithic Node.js application into modular micro-services with dedicated RAG retrieval, ONNX model inference, and real-time Server-Sent Events (SSE) streaming text directly to client interfaces. Live in production with ~150 registered users (55 to 70 active monthly).
- GPU-Accelerated Transcription & Retrieval: Built a Whisper pipeline on ONNX Runtime utilizing lab server GPUs to index a 135-video lecture library, reaching BM25 search retrieval latency of 3 ms with automated fallback to Groq and Anthropic Claude APIs.
- Data Integrity & Booking Engine: Enforced zero double-bookings across calendar reservations using unique compound database indexes in MongoDB.
Full-Stack Deep RL Telemetry & 3D Simulation Platform
React, Three.js WebGL, WebSockets, FastAPI, Python, PyTorch, Qwen3
- Interactive 3D Frontend: Developed a real-time spacecraft visualization dashboard in Three.js and WebSockets, rendering live attitude, thermal, power, and compute states under radiation fault injections.
- Dual-Layer Backend: Built a high-speed FastAPI service bridging the frontend event stream to a POMDP physics simulation environment and deep reinforcement learning agent.
- Model Fine-Tuning: Fine-tuned Qwen3-1.7B with LoRA and GRPO: achieved 87.9% fault recovery vs 79.4% for LSTM-PPO baselines, reducing catastrophic mission failures from 12.4% to 5.9% across 1,500 test episodes (3 random seeds).
- Academic Recognition: Co-authored a research paper under review for AISIGH-SD 2026 (IEEE).
2. Sluice
High-Throughput Event-Driven Transaction Gateway
Node.js, Fastify, Redis, Apache Kafka, PostgreSQL, Docker
- High-Velocity Ingestion: Sustains 2,565 req/s with p50 21.9 ms and zero errors across 102,617 requests under automated k6 load suites.
- Strict Idempotency Guard: Implemented token-bucket rate limiting via atomic Redis Lua scripts and SHA-256 idempotency locks: 10,000 concurrent duplicate requests produce exactly 1 Kafka message and 1 PostgreSQL row.
- Asynchronous Persistence Worker: Designed an at-least-once ledger worker with sorted row locks and offset commits tied strictly to database commits (625 tx/s), with automated Dead-Letter-Queue (DLQ) retry routing.
- Resilience Under Failure: Verified zero request loss during a simulated 30-second PostgreSQL outage under load; Kafka buffered the entire backlog without data corruption.
3. RedisClone
In-Memory Key-Value Datastore & Async Replication Engine in C++17
C++17, Linux epoll, POSIX Sockets, TCP/IP, LLVM libFuzzer
-
Head-to-Head vs Redis 7.0: Reaches 83.4% SET (766k ops/sec) and 80.1% GET (970k ops/sec) throughput on identical 1M-request benchmarks across a 1M random keyspace (
-r 1000000). Sustains 2.49M SET and 2.80M GET ops/sec on single hot-key workloads. -
Reactor Architecture: Single-threaded edge-triggered Linux
epollreactor handling non-blocking TCP streams with pipelined batching. -
Progressive Dual-Table Rehashing: Amortizes hash table expansion incrementally across queries (
$O(1)$ per operation), bounding resize latency jitter to under 1.2 ms. -
Primary-Replica Streaming: Circular ring buffer replication backlog with continuous byte offsets, full snapshot synchronization (
PSYNC ? -1), partial resync (+CONTINUE), andWAITsemi-sync barriers. -
Quantified Failover Durability: Measured write loss by injecting mid-stream
SIGKILLinto the primary: 0.0% loss on LAN, 59% to 62% under 10 ms simulated WAN delay, eliminated completely (0 of 10,000 writes lost) withWAIT 1. -
Fuzzed & Hardened: Fuzzed with LLVM
libFuzzer, ASan, and UBSan across 1.25M+ executions; found and fixed 4 edge-case vulnerabilities, backed by a committed seed corpus.
Deterministic Low-Latency Limit Order Book in Modern C++
C++17, Memory Alignment, Cache-Conscious Data Structures
- Deterministic Trade Matching: Nanosecond-scale trade execution matching orders with strict price-time priority (
FIFO). - Contiguous Memory Structures: Cache-conscious order book layouts minimizing cache misses and dynamic heap allocations during high-frequency order placement, cancellation, and execution.
| Frontend & UI |
|
| Backend & APIs |
|
| Languages & Systems |
|
| Databases & Messaging |
|
| Cloud & DevOps |
|
| Testing & Reliability |
|
- 2nd Prize, CODEX 2026: Solar Irradiance Forecasting competition, organized by Don Bosco College, Bangalore.
- Top 100 out of 30,000+ Participants: Meta PyTorch OpenEnv Hackathon (Scaler School).