Architecting systems,
shipping intelligence.
I am Suhrit Ghimire, a Computer Science undergraduate at DTU building high-performance distributed software, edge architectures, and applied AI systems — bridging first-principles systems engineering with intelligent applications.
My work spans low-latency distributed caching, clean-architecture cross-platform apps, DOM automation engines, and multi-LLM orchestration pipelines — with a relentless focus on performance, algorithmic efficiency, and production reliability.
What I Build
My work lives at the intersection of Scalable Software Engineering and Applied AI Systems. I architect resilient backend and edge runtimes, craft clean frontend and mobile architectures, and deploy intelligent multi-model cascades that serve real users with minimal latency.
Core Toolkit
Engineering Philosophy
Software is only as good as its reliability under load. Whether designing a 5-layer caching hierarchy with promise coalescing to eliminate cache stampedes, structuring clean decoupled layers (BLoC/Cubit) in cross-platform apps, or building heuristic DOM engines that automate complex workflows, I focus on first-principles system design, correctness, and latency optimization.
As a Codeforces Specialist (1,000+ algorithmic problems solved), I treat performance, algorithmic complexity, and edge cases as primary design constraints — writing clean, maintainable, and type-safe code that scales gracefully from prototypes to high-traffic production.
Featured Highlights
Suchak — Civic-Tech Complaint Mapping Platform
Built a nationwide civic complaint engine on Cloudflare Workers featuring a 5-layer caching hierarchy (in-memory LRU with Promise Coalescing, Cloudflare KV, Postgres JSONB fallback, IndexedDB) eliminating cache stampedes. Powered by a sub-20ms spatial GIS engine resolving GPS taps across 6,543 ward boundaries and a 4-stage multi-provider LLM cascade (Groq/Gemini).
WIZA — Intelligent Visa Form Automation Browser Extension
Engineered and shipped a production JavaScript browser extension automating visa form-filling across dynamic single-page applications. Built an intelligent DOM mutation-observer and heuristic field-matching engine achieving 98%+ accurate element detection, cutting manual data-entry time by 80% for real users.
Sanskrit Emotion Dataset & 3-LLM Ensemble Pipeline
First-author research submitted to LREV (Springer) curating the first 17,500-verse emotion dataset. Architected a distributed 3-LLM ensemble orchestration pipeline with automated consensus scoring, exponential backoff, and asynchronous batch processing to reduce evaluation latency by 65%.
Work History
Work History
Research Scholar
Delhi Technological University
- •Curated and expert-validated the first 17,500-verse Sanskrit emotion dataset (9 rasas), architecting a 3-LLM ensemble orchestration pipeline with prompt-driven labeling to ensure accuracy and consistency at scale.
- •Authored a first-author paper on the dataset methodology, submitted to LREV (Springer); currently under peer review.
Product Engineering Intern
WIZA
- •Engineered and shipped a production browser extension (JavaScript) automating visa form-filling across multiple portals via intelligent field-detection, cutting manual data-entry time by 80% for real users.
Software Engineering Intern
Insight Workshop
- •Built a responsive frontend UI (React, HTML5/CSS3) for the LevelUp live-classes platform.
- •Collaborated cross-functionally (Figma, Git/GitHub reviews, daily stand-ups) within a fast-paced agile team to hit tight delivery timelines.
Academics
Delhi Technological University, New Delhi
B.Tech in Computer Science Engineering
DAV Sushil Kedia Vishwa Bharti School, Kathmandu
CBSE Grade XII
Technical Projects
Suchak — Civic-Tech Complaint Mapping Platform
- Built a nationwide civic complaint platform for Nepal on Cloudflare Workers, architecting a 5-layer caching system (in-process LRU/TTL with Promise Coalescing, Cloudflare KV, Postgres JSONB fallback, Browser Cache/IndexedDB) that eliminates cache-stampede failures under concurrent load.
- Built a sub-20ms GIS engine (Turf.js ray-casting, Promise.all parallel scans) resolving GPS taps to 4 elected leaders across 6,543 wards, and a 4-stage, 15-key multi-provider LLM cascade (Groq/Gemini) formalizing informal complaint text.
- Implemented stateless HMAC-SHA-256 session authentication with PBKDF2 password hashing and Cloudflare-edge rate limiting to secure the account-less admin & voting system.
bRial — Multi-Platform Writer’s Media App
- Built a Flutter & Firebase writers’ platform featuring rich-text editing (super_editor), real-time Firestore sync, and optimized database queries for smoother performance.
- Architected a clean-architecture pattern using BLoC/Cubit to decouple presentation, domain, and data layers, reducing unnecessary UI rebuilds and improving app responsiveness.
WIZA — Visa Form Automation Extension
- Engineered and shipped a production browser extension automating visa form-filling across dynamic portals, cutting manual data-entry time by 80% for real users.
Hate Speech Detection — Comparative Transformer Study
- Built a model-validation framework comparing BERT, RoBERTa, and DistilBERT against a BiLSTM+attention baseline; the fine-tuned BERT model reached 92% F1, a 15+ point gain over a TF-IDF+SVM baseline.
- Structured identical training/evaluation pipelines across all models to isolate architecture-level contribution and ensure reproducible, auditable comparisons.
Technical Skills
Languages
CS Fundamentals
Frontend / Mobile
Cloud / Databases
GenAI / LLMs
ML / NLP
Dev & Tooling
Get in Touch
Phone
+91 9625609307