EXHIBIT 01
圧縮アルゴリズム // RAGCompactPy
DEPLOYED ✓Multi-evolutionary memory-compression framework killing RAG token bloat; custom lifecycle scoring (0.4I+0.3U+0.2R+0.1F) runs promotion and eviction at sub-50ms.
( 01 / 06 )巻 · ARCHIVE
PM–001 · CS ENGINEER · DELHI-NCR, IN
第00巻 · 始動
MODE: EN [RAW_ID]CLICK TO MORPH ⇄
PRIYANKARMAJUMDAR
// CATALOG: PM–001•AUTH: PRIYANKAR•プリヤンカル
AI/ML Engineer & Full-Stack Builder — pre-final year CS @ Maharshi Dayanand University.
Currently: ML Engineering @ Sentix Security
PYTHON — PYTORCH — FASTAPI — RAG PIPELINES — LLMOPS — PINECONE — ONNX — COMPUTER VISION — NEXT.JS — REACT — TYPESCRIPT — POSTGRESQL — 限界突破 — ドドド — PYTHON — PYTORCH — FASTAPI — RAG PIPELINES — LLMOPS — PINECONE — ONNX — COMPUTER VISION — NEXT.JS — REACT — TYPESCRIPT — POSTGRESQL — 限界突破 — ドドド —
Vol. 01/第01章 · 領域 [WHAT I BUILD]
[気]
What / I Build
ARCHIVE // DOSSIER METRICS
PM–SPEC–001// 01. CORE DOMAINApplied ML · RAG Pipelines
// 02. INFERENCE BENCHMARKSub-50ms · Token Optimization
// 03. GLOBAL ACCOLADETop 1% AMD Hackathon (#51/4,894)
// 04. CURRENT POSTML Intern @ Sentix Security
// 05. PROBLEM SOLVINGLeetCode (priyankar_majumdar)
[ 限界突破 · DOMAIN MATRIX ]
Builds at the intersection of applied ML and full-stack engineering — RAG pipelines, computer-vision diagnostics, and real-time collaborative web apps. Currently a Machine Learning Engineer Intern at Sentix Security, shipping vector-search infrastructure and LLM inference systems for threat detection. Ranked top 1% globally at the AMD Developer Hackathon and top 7 nationally at Smart India Hackathon.
↳ CORE METHODOLOGY:Production-grade inference · Minimal latency · Clean system design
BasedDelhi-NCR, India
StudyingB.Tech CSE, MDU (2023–2027)
FocusML systems, RAG, full-stack
Vol. 02/第02章 · 経歴 [EXPERIENCE]
[瞬]
Where / I've Worked
SENTIX SECURITY PVT. LTD.
Machine Learning Intern
- Built a Pinecone/Milvus ingestion pipeline cutting threat-detection retrieval latency 30%+.
- Optimized LLM inference with custom prompt-caching + async FastAPI, sub-second response under peak load.
- Automated the ML deploy lifecycle with GCP CI/CD, monitoring, and retraining triggers.
INTERNWARE, IITM JANAKPURI
Technical Executive
- Built responsive MERN landing pages, cutting dev time from 1 week to 3 days via reusable components.
- Streamlined frontend/backend coordination, halving task completion time.
Vol. 03/第03章 · 実績 [PROJECT EXHIBITS]
[撃]
Selected / Case Files
EXHIBIT 02
リアルタイム設計 // CANVASFigma Clone
DEPLOYED ✓Real-time collaborative design canvas with live cursor presence, instant sync, auth, layers, and undo/redo.
EXHIBIT 03
音響ニューラル // MODALAudioCNN
DEPLOYED ✓CNN trained on ESC-50 for real-time environmental sound classification, backed by a serverless Modal + FastAPI inference layer.
EXHIBIT 04
画像診断AI // ONNXLUNG-AI
DEPLOYED ✓Transfer-learning model for lung cancer detection from CT scans with 83.7% test accuracy, a Gradio demo, and ONNX export.
EXHIBIT // LR INT SUPER SAIYAN GOKU
KI LEVEL: 150,000,000
超サイヤ人 · 怒りの一撃
DOKKAN ARCHIVE // LR INT GOKU
ARCHIVE NO. DBZ-NAMEK-01
The Super Saiyan's Final Showdown
Fierce Hand-to-Hand Battle · Angry Super Kamehameha (ばかやろーっ!)
Iconic legendary battle visual capturing the explosive anger on Planet Namek. Grounded into the dossier as a testament to peak engineering endurance, limit-breaking compute, and high-energy full-stack execution.
Vol. 04/第04章 · 栄誉 [ACCOLADES & ARCHIVE]
[極]
On / The Record
#51 of 4,894 teams worldwide (top 1%) — co-built a token-efficient LLM routing agent, 89.5% accuracy inside a strict 5,333-token cap (2026).
7th of 50 teams, North India regional — advanced to nationals for building real-time mission-critical software solutions.
Solved hundreds of algorithmic challenges across Dynamic Programming, Graphs, and Advanced Trees. Profile: leetcode.com/u/priyankar_majumdar ↗
Engineered with editorial catalog numbering, Japanese cipher animations, and 16-bit Ki surge renderers inspired by classic Shonen & vinyl micro-archives.
Vol. 05/第05章 · 兵装 [AI/ML & DEV SPECS]
[破]
The / Toolkit
Vol. 06/第06章 · 交信 [TRANSMISSION & CONTACT]
[電]
Let's / Talk
Open to ML engineering and full-stack roles. Reach out for collaborations, discussions on RAG systems, or high-performance engineering.