002 · About
About
I'm a CS engineer from IIIT Agartala who works across the whole product: deciding what to build, designing how it feels, engineering how it scales, and explaining it so others can build on it.
Product managementDirectionDecides what is worth building before anything is built.
- Product Management Intern, Darwix AI: conversational AI for the BFSI sector
- Core Team Member, Entrepreneurship Cell (E-Cell), NIT Agartala
Full-stack & systemsStructureBuilds systems that fail gracefully.
UI/UX designFormMakes the complicated feel obvious.
- UI/UX Trainee, Techugo: wireframes and interactive Figma prototypes
- Design Team Lead, Developers & Coders Club (DCC), NIT Agartala
- This site: designed and built end to end
Writing & researchRecordWrites the manual so others can build on it.
TeachingTransmissionExplains it until it clicks.
- Vice-President, CODE, IIIT Agartala
- Design Team Lead, DCC, NIT Agartala
| Class of 2026 | IIIT Agartala |
|---|---|
| PM Intern | Darwix AI |
| UI/UX Trainee | Techugo |
| LeetCode Knight | Top 3.28% |
| 2 flagship systems | LogPort · GraphSTL |
| Base | India · IST |
003 · Work
Work
001● Shipped
LogPort
LogPort · Real-time log ingestion pipeline
━ each log, on the request path━ batches, off the request path┅ failed → DLQ - Problem
- Take in logs at volume without losing them when the database fails, and show them live as they arrive.
- Stack
- ReactNode.js · FastifyRedisClickHouseWebSocketDocker
Highlights
- Ingestion: a Fastify API validates, authenticates and rate-limits batched writes, then LPUSHes each log onto a Redis list; a separate worker drains it with LMPOP in batches of 100 into ClickHouse.
- Fault tolerance: failed inserts retry with exponential backoff (500 → 1000 → 2000 ms), then move to a capped Redis dead-letter queue that can be replayed.
- Live and secured: the API publishes every accepted log over Redis Pub/Sub to a WebSocket feed with auto-reconnect and pause/resume; JWT or API-key auth; a Redis sorted-set limiter per key and endpoint.
002● Shipped2026
GraphSTL
GraphSTL · Header-only C++17 graph library
#include <GraphSTL/graph.h>#include <GraphSTL/graph_algorithms.h>using namespace custom_stl; graph<char, int, true> g; // directedfor (char c : {'A','B','C','D','E','F'}) g.addNode(c);g.addEdge('A','B',4); g.addEdge('A','C',2);g.addEdge('B','D',10); g.addEdge('C','E',3);g.addEdge('C','D',5); g.addEdge('D','F',1);g.addEdge('E','F',5); auto [cost, path] = dijkstra(g, 'A', 'F');// cost = 8, path = A C D F- Problem
- Give C++ a graph container that feels like the standard library, with algorithms decoupled from the container.
- Stack
- C++17CMakevcpkgGitHub Actions
- 12
- Graph algorithms
- 46
- Unit tests in CI
- 0
- Dependencies
Highlights
- graph<T, Weight, IsDirected> with O(1) average-case node lookup (hash maps both ways between node data and IDs) and STL-style forward iterators.
- 12 algorithms as free functions: BFS, DFS, Dijkstra (heap and naive), Bellman-Ford, A*, Floyd-Warshall, Kruskal's MST, Kahn's topological sort, Kosaraju's and Tarjan's SCC, plus cycle detection.
- 46 unit tests run in GitHub Actions CI, plus a benchmark comparing heap-based Dijkstra, O((V+E) log V), with the O(V²) linear scan.
004 · Skills
Skills
Core
- C++Used in · GraphSTL
- TypeScript / JS
- React
- Node.jsUsed in · LogPort · Fastify API
- RedisUsed in · LogPort · queue, Pub/Sub, rate limiter
- ClickHouseUsed in · LogPort · storage
- FigmaUsed in · Techugo · DCC Design Lead
- ProductUsed in · Darwix AI · E-Cell
All skills
- Languages
- CC++JavaScriptTypeScript
- Technologies
- HTML5CSS3Tailwind CSSReactNode.jsFastifyWebSocket
- Databases
- MongoDBSQLRedisClickHouse
- Fundamentals
- Data structures & algorithmsOOPOperating systemsDBMSDistributed systems
- Tools
- GitGitHubVS CodeDockerCMakeFigmaCanvaFilmoraAdobe Acrobat
Certifications
Numbers
- 1961
- LeetCode rating · Knight
- 3.28%
- Global top, LeetCode
- 305
- Problems solved, LeetCode
- 39,725
- JEE Main 2022 AIR
005 · Contact