GeezerVison is showing github.com — manaskng - Overview
Pre-final Year CSE DTU'27
Knight(2051) @leetcode
DS Intern@VLED LAB
Computer Engineering student at Delhi Technological University (DTU) — CGPA 9.13.
I build high-performance software across systems programming, distributed systems, AI, and full-stack engineering. Recent work spans a zero-dependency DPI engine at 438K pkt/s, a RAG-powered collaborative developer platform, an async ML opportunity aggregator, and a recursive-descent compiler.
Current focus
- Systems — DPI, packet analysis, low-level C++17
- Agentic AI — RAG pipelines, vector search, LLM orchestration
- Compilers — parsing, IR, CFG optimizations
- Distributed systems — microservices, async ETL, caching layers
| 🛡️ PacketSentinel |
🔗 Relay |
|
Multi-threaded Deep Packet Inspection & Malware Detection Engine
- 438K pkt/s via lock-free pipeline (FNV-1a sharding), zero mutex contention on the hot path
- Byte-level TLS SNI / HTTP Host / DNS parsing via
std::string_view, zero heap allocs
- Shannon entropy + Welford variance anomaly in O(1) memory; Random Forest 97% precision
- 1.7× speedup over baseline; classifies 15+ app types from live traffic
C++17 · Python · scikit-learn · libpcap · Scapy · CMake
Code · Live
|
Adaptive RAG-Powered Collaborative Developer Ecosystem
- Custom RAG: 768-dim Gemini embeddings + Atlas Vector Search, sub-50ms retrieval across 10K+ docs (60× over regex)
- Real-time collab editor (Socket.IO), 11 compiled languages, dual-compiler failover, 50+ rooms
- Redis sliding-window rate limiter + dual Gemini model failover
- 40+ REST endpoints, Dockerized CI/CD, MongoDB TTL auto-purge
MERN · Socket.IO · Redis · Gemini API · Docker · Vector Search
Code · Live
|
| 🔍 OppLens |
⚙️ AstraC Compiler |
|
Distributed Opportunity Aggregator & ML Recommendation Engine
- Fault-tolerant Playwright ETL across 5 platforms, 5K+ opps/week; composite-key hashing cuts writes ~95%
- Atlas Vector Search (HNSW): O(N) → O(log N), ~10× speedup; sentence-transformers dedup
- Hybrid recommender — TF-IDF + Collaborative Filtering + Vector Search in a FastAPI microservice
- Redis cache: p95 inference 400ms → <80ms (80% reduction)
Next.js · FastAPI · Python · sentence-transformers · Redis · Playwright
Code · Live
|
C-Subset Recursive-Descent Compiler
- Hand-written recursive-descent parser → AST → semantic analysis (type checking, scope resolution)
- 3-address IR (TAC) with basic-block CFG — constant folding, DCE, copy propagation → ~25% fewer instructions
- 100+ pytest cases: pointer arithmetic, nested scopes, register spilling
- Symbol-table-driven linker helper across translation units
C++17 · CMake · pytest
Code
|
📊 GitHub & Competitive Programming Stats
🏆 Competitive Programming & Achievements
|
|
|
| 🥇 |
Goldman Sachs India Hackathon 2026 |
Rank 1 · 100/100 · 15,000+ submissions · Recursive JSON→TypeScript parser in C++ |
| ⚔️ |
LeetCode Knight |
Top 2% · Rating 2051 · 800+ problems · Rank 307/35,237 (Biweekly 174) |
| 📘 |
AlgoUniversity DP Camp |
Top 2.5% of 40,000 · Codeforces Master mentorship |
| 🎓 |
SDE Intern @ IIT Ropar |
Vicharanashala Lab · 2 merged PRs · Setup pipeline + UI bug fix |
| 🌟 |
GDG Solution Challenge 2025 |
Certificate of Achievement · AI-driven solution |
| 🔖 |
GitHub Badges |
Pull Shark · Starstruck |
Open to SDE internships, research collaborations, and open-source contributions.
[email protected]
Pinned
Loading
-
Real-time collaborative developer workspace featuring a workspace-wide RAG AI agent driven by Atlas Vector Search
JavaScript
25
-
Distributed scraping and semantic deduplication engine using Sentence Transformers and Atlas Vector Search
TypeScript
48
-
multi-stage compiler for a C subset in C++: hand-written recursive-descent parser, AST construction, semantic analysis (type checking, scope resolution), and x86-64 code generation.
Makefile
2
-
multi-threaded Deep Packet Inspection (DPI) engine written in pure C++17. Classifies live network traffic via TLS SNI extraction at wire-speed, featuring a real-time analytics dashboard
C++
2
-
-
Applied machine learning pipeline utilizing the aeon framework to benchmark Proximity Forest for scalable time-series classificatio
Jupyter Notebook
Uh oh!
There was an error while loading. Please reload this page.
|