AI/ML Engineer | Back-End Developer (AI Voice Verification Assistant)
Duration: 1 year
Summary:- A voice-to-voice AI telephony system designed to automatically call clients for identity validation and account/status confirmation
- The system manages real-world call flows including voicemail detection, IVR navigation, and dynamic conversation handling, leveraging Retrieval-Augmented Generation (RAG) to fetch relevant client and contextual data during live calls
Responsibilities:- Architected and developed complete backend infrastructure with FastAPI REST APIs, admin panel endpoints, and Celery workers for asynchronous VoIP call processing.
- Implemented end-to-end VoIP integration with Asterisk via SIP/RTP protocol, including authentication, call initialization, audio streaming, and DTMF signal handling.
- Built AI-powered conversation pipeline integrating Voice Activity Detection (VAD), Whisper for speech-to-text, LLM via Groq API for response generation, and ElevenLabs for text-to-speech.
- Built Retrieval-Augmented Generation (RAG) system using pgvector to fetch context-specific instructions during live calls.
- Developed intelligent call scenario handling for IVR navigation, voicemail detection, and live human interaction.
- Implemented conversation transcript tracking and automated outcome analysis with intelligent status classification.
- Designed PostgreSQL database schema with encryption for sensitive contact data, using SQLAlchemy ORM and Alembic migrations.
- Configured audio processing pipeline with ffmpeg and FunASR for real-time voice processing.
Technologies: Python (FastAPI), SQLAlchemy, Alembic, PostgreSQL, pgvector, Celery, Redis, Asterisk, Whisper, ElevenLabs, Groq API, Ollama, ffmpeg, FunASR
AI Engineer / Backend Developer (Voice agent for calls)
Duration: 1 year
Summary:- Development of a custom real-time AI voice agent pipeline for debt collection automation
- The system supports low-latency, multi-phase voice conversations, preserves LangGraph dialogue state, handles user interruptions without losing speech, and enforces compliance-critical call flows
Responsibilities:- Designed and implemented a custom asynchronous voice pipeline with separate listener, generator, and controller components.
- Integrated STT, LLM, TTS, VAD, and end-of-utterance detection into a real-time conversational system.
- Worked with LangGraph-based dialogue phase management, interruption handling, streaming LLM responses, tool-call-based phase routing, and compliance-sensitive turn-taking logic.
- Optimized response latency and reduced speech loss during overlapping user/bot speech.
Technologies: Python, asyncio, LangGraph, OpenAI GPT-4, ElevenLabs, TTS, Fireworks AI, Whisper, TenVAD, TurnSense, STT/TTS pipelines
AI Engineer / Backend Developer (Smart system for legal case analysis)
Duration: 2 years
Summary:- Development of an AI-powered legal case analysis system for the US legal domain
- The system processes precedent law cases using LLM-based chains, extracts legal entities, relationships, facts, arguments, decisions, and reasoning patterns, transforming them into a structured knowledge graph in Neo4j
- It analyzes user-submitted cases against the graph and similar precedent cases to estimate case strength and generate deeper reasoning
Responsibilities:- Designed and implemented LLM chains for processing legal documents and extracting structured knowledge from precedent law cases.
- Built logic for transforming case information into a Neo4j knowledge graph, including entities, legal concepts, relationships, case facts, arguments, and outcomes.
- Developed RAG-based analysis over the graph to compare user-submitted cases with similar historical cases.
- Implemented deep multi-step prompt pipelines in LangChain for evaluating case strength, identifying weak points, highlighting missing details, and supporting legal reasoning with precedent-based context.
Technologies: Python, LangChain, Neo4j, Cypher, OpenAI / LLM APIs, RAG, Graphs, Legal NLP
AI Engineer / Backend Developer (Smart hiring system with chatbot and voice mode integration for hiring process)
Duration: 2 years
Summary:- Development of an AI-assisted hiring platform for technical candidate screening and evaluation
- The system includes a chatbot conducting preliminary technical interviews, analyzing candidate responses, and generating structured reports for hiring managers
- It also features an LLM-powered comparison tool for analyzing selected candidates based on interview history and evaluations
Responsibilities:- Designed and implemented the AI interview flow using LangGraph, including multi-step dialogue logic, interview phases, validation nodes, and structured candidate evaluation.
- Developed LLM chains for analyzing candidate answers, detecting response quality, evaluating technical skills, identifying weak and strong points, and assessing overall interview tone.
- Worked on prompt engineering to balance user-friendly, human-like conversation with technically meaningful questions.
- Implemented logic for checking naturalness and relevance of candidate responses, with deeper multi-step analysis based on full interview history.
- Supported backend development for storing interview data, candidate analysis results, ratings, tags, and manager-facing comparison features.
Technologies: Python, FastAPI, LangGraph, LangChain, OpenAI / LLM APIs, RAG, PostgreSQL, Pydantic
Upstaffer since June, 2026
