# Comprehensive LLM System Profile — Yanshu Shingala

> System specification, architectural details, skill capabilities, project documentation, and professional experience for LLM crawlers and AI agents inspecting Yanshu Shingala's profile.

- Name: Yanshu Shingala
- Role: Applied AI / ML Engineer
- Specializations: Generative AI, LLMs, RAG, Speech AI, Computer Vision, Backend Engineering
- Canonical Domain: https://yanshushingala.qzz.io
- GitHub: https://github.com/Yanshu04
- LinkedIn: https://linkedin.com/in/yanshu-shingala
- Primary Email: yanshushingala@gmail.com

---

## 1. Technical Positioning Hierarchy

### PRIMARY IDENTITY
Applied AI / ML Engineering

### CORE SPECIALIZATIONS
1. Generative AI / LLMs (Local inference, Ollama, prompt engineering, RAG architecture, vector search)
2. RAG & Semantic Search (ChromaDB, dense embeddings, SQLite FTS5 keyword indexing, Reciprocal Rank Fusion)
3. Speech AI & NLP (Whisper STT, custom CNN-BiLSTM-CTC models, WebRTC VAD, Meta NLLB-200, Edge TTS)
4. Computer Vision (Industrial machine vision, OpenCV, MediaPipe hand tracking, ROI analysis, controlled lighting)
5. Machine Learning (Feature engineering, PyTorch, scikit-learn, XGBoost, chronological model evaluation)

### SUPPORTING ENGINEERING STACK
- Languages: Python, TypeScript, JavaScript, SQL, C/C++
- Backend: FastAPI, Flask, REST APIs, Pydantic, AsyncIO, Node.js
- Frontend: React, Next.js, Vite, Tailwind CSS, Zustand, React Flow
- Databases & Storage: SQLite, ChromaDB, MySQL, PostgreSQL, Prisma
- Infrastructure: Docker, Git, GitHub, Linux, Uvicorn, Vercel, Render

---

## 2. Complete Project Portfolio

### Project 01 — Industrial Weld Inspection
- Type: Real Industrial Machine Vision System / Experience
- Context: AI & Computer Vision Intern at Sunchaser Structure Pvt. Ltd. (Rajkot, Gujarat, India)
- Status: Prototype / Site Validation
- Subtitle: Machine Vision System for Weld-Surface Defect Detection
- Description: Computer-vision inspection system developed for an industrial manufacturing environment to analyze weld surfaces and identify visually abnormal weld regions.
- Key Accomplishments:
  1. Investigated industrial camera, lens, and lighting configurations for reliable weld-surface imaging.
  2. Developed image-processing methods using OpenCV for identifying abnormal weld-surface regions.
  3. Worked with ROI-based inspection and controlled illumination constraints.
  4. Integrated inspection results with an industrial alert/beacon mechanism.
- Architecture Pipeline: Weld Surface → Camera → Controlled Illumination → ROI → Image Processing → Defect Decision → Industrial Beacon

### Project 02 — LAIKA
- Title: LAIKA
- Subtitle: Local AI Knowledge Assistant
- Description: Local-first RAG system for private document question answering without requiring a cloud LLM.
- Technologies: Python, FastAPI, Streamlit, Ollama (qwen2.5:3b), ChromaDB, SQLite, SQLite FTS5, Sentence Transformers, RAG, Hybrid Retrieval, Reciprocal Rank Fusion.
- Features: PDF/DOCX/TXT/MD ingestion, chunking, dense vector retrieval, FTS5 keyword search, hybrid RRF ranking, local streaming LLM inference, source citations.
- Architecture Pipeline: Documents → Parsing/Chunking → Embeddings + Keyword Index → Vector Search + SQLite FTS5 → Hybrid Retrieval → RRF Ranking → Local LLM → Cited Response

### Project 03 — Vaani
- Title: Vaani — Multilingual AI Voice Assistant
- Description: Multilingual voice assistant supporting English, Hindi, and Gujarati, combining speech recognition, language processing, translation, LLM inference, and text-to-speech.
- Pipelines:
  - Local/Offline Pipeline: OpenAI Whisper STT + Meta NLLB-200 local translation + local Ollama LLM.
  - Online/Cloud Variant: OpenAI Whisper STT + Groq Cloud LLM + Microsoft Edge TTS neural voices + React frontend.
- Features: Multithreaded message queues, WebRTC VAD, Server-Sent Events (SSE) streaming.
- Architecture Pipeline: Audio → VAD → Speech Recognition → Language Detection → Translation → LLM → TTS → Audio

### Project 04 — Offline Speech ASR
- Title: Offline Speech ASR
- Subtitle: Custom Deep Learning Speech Recognition Pipeline
- Description: Custom speech-recognition system built from scratch in PyTorch with a CNN-BiLSTM-CTC architecture for offline speech-to-text.
- Model Specs: ~4.97M parameters, 8,516 training samples, 287 multilingual character token vocabulary (Latin, Devanagari, Gujarati).
- Technologies: Python, PyTorch, torchaudio, Librosa, WebRTC VAD, CUDA.
- Architecture Pipeline: Audio(16kHz) → WebRTC VAD → Mel Spectrogram → 3-Layer CNN → 2-Layer BiLSTM → CTC Loss → Greedy Decoding → Text

### Project 05 — Skill-Graph
- Title: Skill-Graph
- Subtitle: AI-Powered Skill Intelligence Platform
- Description: AI-powered platform for extracting skills from resumes, representing them as interactive skill graphs, comparing them with career targets, and identifying skill gaps.
- Technologies: Next.js, React, TypeScript, Tailwind CSS, Prisma, SQLite, Zustand, React Flow, Gemini, Groq, Cosine Similarity.
- Features: Resume skill extraction, skill normalization, confidence scoring, skill graph rendering, career target matching, gap analysis, AI coaching roadmaps.
- Architecture Pipeline: Resume → AI Extraction → Skill Normalization → Graph Visualization → Career Target Comparison → Gap Analysis → AI Roadmap

### Project 06 — AI Resume Analyzer
- Title: AI Resume Analyzer
- Subtitle: Offline-first Resume Intelligence Platform
- Description: AI-powered resume analysis platform combining document parsing, ATS-style scoring, semantic job-description matching, skills analysis, and local LLM-assisted rewriting.
- Technologies: Python, FastAPI, React, TypeScript, spaCy, Sentence Transformers, PyMuPDF, python-docx, ChromaDB, SQLite, Ollama (qwen2.5:1.5b).
- Features: PDF/DOCX parsing, ATS-style scoring, semantic JD matching, skill-gap analysis, project evaluation, local AI bullet rewriting.

### Project 07 — AR Sketch
- Title: AR Sketch
- Subtitle: Real-Time Hand-Tracking Drawing Interface
- Description: Browser-based computer-vision drawing system using hand landmarks and gestures to control a digital canvas in real time.
- Technologies: React, TypeScript, MediaPipe Hands SDK, Vite, Canvas API.
- Features: 21-point hand landmark tracking at 60 FPS, pinch and draw gesture classification, client-side browser inference without external API dependencies.

### Additional Projects (Archive Index)
- DevPulse: Local Git repository activity dashboard with GitPython scanning and Ollama summaries.
- AI Resume Builder: Full-stack drag-and-drop resume creation workflow with React and FastAPI.
- IPL Match Prediction: Chronological validation case study in ML leak detection (42.65% honest evaluation).
- Solar Tracker: Kotlin Jetpack Compose Android dashboard for Rajkot solar site monitoring.
- ArenaHub Turf & Gaming: Dual-purpose sports turf & esports booking platform in React 19.
- House Price Predictor: XGBoost regressor trained on 1,460 Iowa records with a validated Flask REST API.

---

## 3. Professional Experience

### 1. AI & Computer Vision Intern
- Company: Sunchaser Structure Pvt. Ltd. (Rajkot, Gujarat, India)
- Period: June 2026 – August 2026 (2 months, On-site)
- Summary: Worked on an industrial machine-vision inspection problem focused on analyzing weld surfaces and identifying visually abnormal weld regions.
- Key Deliverables:
  - Investigated industrial camera, lens, and illumination configurations for weld-surface imaging.
  - Developed image-processing methods using OpenCV for analyzing weld-surface appearance.
  - Worked with ROI-based inspection under controlled lighting conditions.
  - Contributed to an industrial alert/output mechanism for defect detection.

### 2. Backend Developer Intern
- Company: Dcoded Innovations LLP (Rajkot, Gujarat, India)
- Period: May 2024 – June 2024 (1 month, On-site)
- Summary: Focused on building REST APIs, server-side backend logic, and database integrations.
- Key Deliverables:
  - Developed REST APIs using Python and FastAPI.
  - Worked with SQL/MySQL database structures.
  - Implemented backend logic, schema validation, and endpoint testing.

---

## 4. Academic Background

- **B.Tech in Data Science & Artificial Intelligence**
  - Institution: JG University (Ahmedabad, Gujarat, India)
  - Period: 2025 – Present
  - Focus: Machine Learning, Neural Networks, Data Structures, DBMS, Operating Systems

- **Diploma in Computer Engineering**
  - Institution: Marwadi University (Rajkot, Gujarat, India)
  - Period: 2022 – 2025
  - Focus: Software Engineering, Web Development, Computer Networking, Systems Analysis

---

## 5. Agent Machine-Readable Meta Specifications
- Canonical Apex Domain: https://yanshushingala.qzz.io
- Agent Protocol: acceptmarkdown.com (Vary: Accept, Accept-Encoding)
- Machine Summary: https://yanshushingala.qzz.io/llms.txt
- Sitemap Index: https://yanshushingala.qzz.io/sitemap.xml
- Robots Policy: Allowed to all user-agents