DWG. 002 / PROJECT SHEETS

Five systems, drawn to completion.

Each one runs as a complete system. Four are my own builds, one comes from my internship at Medha Tech Solutions.

DEPLOYED FOR DELIVERAIL

EMMA, AI Voice Receptionist

A voice agent that handles reception calls end to end, fully on device, for a logistics company.

Speech to text with faster-whisper, reasoning through a local Ollama LLM, speech out with Coqui TTS.
Request validation and a knowledge grounding check run in plain Python before the LLM is invoked.
The model can only answer from RAG grounded knowledge or trigger a live shipment booking through tool calls.

TESTED AT CMR UNIVERSITY

A private, fully offline AI assistant where every assistant has its own knowledge base, prompt and settings.

Answers from PDF, DOCX, TXT, CSV and Excel with streamed, cited responses and a confidence score.
Hybrid retrieval combining FAISS and keyword search with reranking, plus a voice interface.
CSV and Excel analytics with KPI reporting, user accounts, and backup and restore.

HUMAN APPROVAL FIRST

A personal AI command center that runs my job search across ten live product sections.

Syncs and classifies job emails with the Gmail API, pulls roles from Adzuna and scores matches against my resume.
Tracks every application through a pipeline, interview prep and recruiter follow ups.
Every AI suggested action needs explicit confirmation before anything is sent or written. No scraping or ToS violating automation.

RUNNING IN PRODUCTION

An AI streamer persona that replies live to real comments and messages on Instagram and Discord.

An LLM backed conversational pipeline that runs continuously in production.
Reliability work in progress: retry logic for failed replies and a supervisor process.
Growing the evaluation set toward 50 or more test cases.

MEDHA TECH SOLUTIONS, INTERNSHIP

A RAG chatbot that answers from an organization’s own knowledge base and stays inside its domain.

SHEET 05 / 05Internship work, source private