What I've built

My Projects

Production systems spanning AI agent design, backend development, voice technology, and full-stack deployment. Every project is built with a focus on reliability and real-world impact.

Production Work

@ Telezer Technologies
verified
LangChain LangGraph Faster-Whisper Qwen 7B SQLite

QC AI Agent

Production-grade Quality Control agent for automated call evaluation. Built the full pipeline: call recordings → Faster-Whisper STT → Qwen 7B local LLM → structured quality scoring and business insights. Implemented multi-step reasoning using LangGraph with fallback logic deployed on on-premise Linux servers.

check_circle Live in Production dns On-Premise Linux psychology Multi-step Reasoning
call
WebRTC · LiveKit VICIdial Asterisk STT · TTS

Real-Time Voice AI Pipeline

End-to-end voice AI system for BPO automation, integrating VICIdial and Asterisk telephony with WebRTC (LiveKit) for real-time AI voice interactions. Designed low-latency conversational AI using ChatGPT and local LLMs with STT and TTS — managing complete call flows from ingestion to agent response.

check_circle Live in Production speed Low-Latency mic Real-time Voice

Personal Projects

account_tree
Python Ollama LangChain ChromaDB FAISS

Custom AI Agent System

Fully local AI agent framework implementing the complete agent loop: intent parsing → tool selection → multi-step reasoning → structured output. Built RAG pipeline with ChromaDB and FAISS for context-aware responses over private data. Implemented tool-calling with JSON schema validation, fallback handling, and memory management.

lock 100% Local memory RAG + Vector DB
health_and_safety
Node.js Express.js MongoDB Render Netlify

Health Query AI Platform

Full-stack AI-powered health query platform using LLM-based responses with safety guardrails and domain-specific prompt engineering. Deployed RESTful APIs on Render with CI/CD frontend on Netlify. Focused on response reliability using structured prompts and context management.

cloud Deployed Live security Safety Guardrails
sentiment_neutral
Python NLTK SpeechRecognition NLP

Speech-Based Mood Detection

End-to-end pipeline from audio ingestion → speech-to-text → NLP → emotion classification. Implemented multi-class emotion detection using vocal patterns and textual features. Same core STT/NLP architecture later applied to production voice AI systems at Telezer Technologies.

mic Audio → Emotion school OCAC Training

Impact Metrics

3+

Personal Projects

2

Production Systems

<250ms

Voice Latency Target

smart_toy

Talk to Virtual Me

Ask me anything about my projects, skills, or experience.

Start Chat →