>_ abd.dev
// rasterizing 3d ascii marlin...

shaik abdullah

AI/ML Engineer & Systems Builder.
Specializing in RAG Systems & Agentic AI.

// Computer Science @ CBIT Hyderabad. Building first-principles AI agents, hybrid retrieval pipelines (BM25 + Dense + FlashRank), and autonomous real-time systems.

Open for AI/ML engineering roles, RAG research & collaborations →

Bridging first-principles
AI reasoning with robust
software engineering.

I'm a Computer Science student at CBIT Hyderabad focusing on Agentic AI, RAG architectures, and real-time intelligent systems. I believe in understanding how LLMs reason without hiding behind bloated framework abstractions.

01
First-Principles Agent Loops. Understanding prompt construction, tool-invocation dispatch, and state memory from scratch.
02
Hybrid Retrieval Over Naive Vector Search. Combining BM25 keyword matching with dense embeddings and FlashRank cross-encoder reranking to eliminate hallucinations.
03
Real-Time Signal & Vision Systems. Extracting actionable acoustic features and deploying edge vision models for autonomous drone disaster response.
04
Open Source & Community Leadership. AI/ML Co-Lead at CBIT DSC and Vice Head at Chaitanya ASTRA, coordinating national-level tech events and open-source contributions.

Featured systems
& engineering projects.

// Core open-source projects across Agentic AI, RAG architectures, and speech processing systems.

01
AI Agent from Scratch (No Frameworks)
Custom AI agent built in pure Python without LangChain or agent frameworks, using the Sarvam AI API for autonomous reasoning.
  • Implemented the full agent loop manually: prompt construction, response parsing, and tool-invocation logic with live web search capability.
  • Demonstrated core agentic concepts (reasoning, tool-calling, action-taking) from first principles without prebuilt framework abstractions.
#Python#Sarvam AI#Agentic AI#Tool Calling#No-Framework
02
Financial RAG Pipeline
End-to-end Financial RAG pipeline to ingest, index, and query complex financial documents with hybrid retrieval and reranking.
  • Built with LangChain, ChromaDB vector store, Sentence Transformers embeddings, and the Groq API for high-speed inference.
  • Designed a multi-stage hybrid retrieval layer combining semantic dense search, BM25 sparse retrieval, and FlashRank cross-encoder reranking to eliminate hallucinations.
#LangChain#ChromaDB#BM25#Groq API#FlashRank#RAG
03
AI-based Mock Interview Platform
Data-driven technical interview platform generating role-specific questions with adaptive follow-up logic and speech signal evaluation.
  • Generates and evaluates 500+ role-specific questions with dynamic context-aware follow-up questioning.
  • Built a speech/voice processing pipeline extracting acoustic features (pitch variance, pause distribution, confidence) to generate automated feedback reports.
#Python#Speech AI#FastAPI#OpenCV#Feature Extraction
04
Juno SDK (Open Source Contribution)
Contributed to Juno, an open-source SDK for building developer applications; had a merge request accepted and merged into the upstream codebase.
  • Identified and resolved core dispatch handler typing inconsistencies.
  • Merge request accepted and integrated into the primary open-source release.
#Open Source#TypeScript#SDK Architecture

Tools I use to build
intelligent systems.

// Technical competencies across AI/ML engineering, RAG pipelines, backend architectures, and data tooling.

AI & LLM Systems
RAG Pipelines & Hybrid SearchCore AI
BM25 + Dense Embeddings + RerankingRetrieval
Agentic AI & Tool-Calling LoopsReasoning
LangChain & ChromaDB (Vector DB)Orchestration
Prompt Engineering & TransformersLLM
Languages & Backend
PythonPrimary
JavaScript & Node.jsFull-Stack
FastAPI & FlaskAI APIs
Express.js & REST APIsBackend
WebSockets & CReal-Time
ML, Data & Vision
NumPy & scikit-learnML / Data
TensorFlow & Deep LearningModels
OpenCV & Signal AnalysisVision / Audio
StreamlitRapid Prototyping
Databases & Tools
ChromaDB (Vector Database)Embeddings
MySQL & MongoDBStorage
Git & GitHubVCS
Docker & LinuxDevOps

Latest GitHub Commits.

// Last 5 public git commits synchronized in real-time from github.com/sheikhabd22

commit a4f8e21 [Financial_rag:main]
Recent

Implemented hybrid retrieval layer combining BM25, semantic vector search and FlashRank reranking.

commit b7d19c4 [Ai_mock_interview:main]
Recent

Added real-time acoustic pitch variance and pause feature extraction pipeline.

commit e9c32f0 [ai-agent-scratch:main]
Recent

Built core tool-invocation and live web search execution loop without external agent frameworks.

commit f2a10d8 [juno-sdk:upstream/main]
Recent

Open source contribution: resolved SDK core dispatch typing and merge request merged.

commit d3e55b1 [abddev:main]
Recent

Calibrated retro dark ponytail portfolio with 3D ASCII GLTF marlin rasterizer and live GitHub sync.

Writing & thinking
out loud.

// In-depth technical articles on Agentic AI, RAG architectures, and systems engineering. Click any post to read the full article.

01
Building an AI Agent from Scratch Without Frameworks[ read full post ↗ ]
Why I built an autonomous tool-calling agent in pure Python without LangChain or LangGraph — demystifying the core agent reasoning loop.
02
Hybrid Retrieval in Financial RAG: BM25 + Embeddings + Reranking[ read full post ↗ ]
How combining dense semantic vectors with sparse lexical BM25 and FlashRank cross-encoders eliminates hallucinations in financial document QA.
03
Feature Extraction in AI Mock Interviews: Speech, Pitch & Confidence[ read full post ↗ ]
Analyzing acoustic audio features alongside LLM transcript evaluation to generate holistic communication feedback reports.
04
Autonomous Drone Navigation in Disaster Response (NIDAR Competition)[ read full post ↗ ]
Developing mission planning logic and real-time sensor ingestion for Team ASTRA's dual-drone search-and-rescue system — Top 10 finish among 600+ teams.

Run commands. Get
instant answers.

// Execute terminal commands directly inside the web interface.

bash — 80×24UTF-8
visitor@abd-dev:~$welcome
Welcome to shaik-abdullah.dev CLI v1.0.0. Type 'help' to list available commands.
visitor@abd-dev:~$

Send a payload.

// Inquire about AI/ML engineering roles, RAG consulting, or project collaborations.