
AI Engineer with 1+ years in Production Voice AI & LLM Fine-tuning
AI is analyzing your overall score…
Identifying your key strengths…
Evaluating your skill match against the job requirements…
Assessing your cultural and operational fit
AI/ML Engineer specializing in production Voice AI, LLM fine-tuning, retrieval-augmented systems, and open-source AI development. Combines deep learning research with full-stack engineering to ship end-to-end AI products from prototype to deployment. Passionate about bridging cutting-edge ML research with practical engineering to drive real business impact.
Anand Institute of Higher Technology
B.Tech · Artificial Intelligence and Data Science
August 1, 2022 – June 30, 2026
F22 Labs
AI Engineer Intern
December 1, 2025 – Present
Chennai, Tamil Nadu, India
Shiash Pvt Ltd
Data Science Intern
July 1, 2025 – November 1, 2025
Chennai, Tamil Nadu, India
UptoSkills
Data Analytics Intern
January 1, 2025 – April 1, 2025
India
Arul Technologies Pvt Ltd
AI/ML Intern
November 1, 2024 – December 1, 2024
Chennai, Tamil Nadu, India
TTS Fine-Tuning & Task Arithmetic Research
January 1, 2026 – Present
Pioneered Task Arithmetic for TTS, combined fine-tuned female voice + Indian accent Kokoro models in shared weight space at a=0.6, β=1.0 without any retraining; achieved MOS 4.4 and 55% listener preference vs 27% baseline. [Blog] Fine-tuned XTTS-v2 GPT component (DDP across 2 GPUs, 30 epochs, 500 synthetic clips) reduced WER by 58.4% (18.54% 7.71%), semantic similarity +12.1%, NISQA MOS +5.5%. Fine-tuned Kokoro-82M on 4,358 Indian-English audio clips (2-stage StyleTTS2) with custom Indian G2P phoneme lexicon; improved Indian proper noun pronunciation from 3.4/10 to 8.8/10.
View ProjectReal-Time Multilingual Translation System
January 1, 2026 – Present
Architected browser-native live speech-to-speech translation system across 5+ Indian languages (Hindi, Tamil, Telugu, Kannada, Malayalam, Bengali) at ~380ms E2E latency supporting 25+ concurrent listeners per room. Reduced cross-lingual TTS latency by 83% (650ms → 75ms) through systematic 5-provider benchmarking; engineered 3-stage pipeline: Deepgram Nova-3 STT (~150ms) → Sarvam Translate (~45ms) → ElevenLabs Flash v2.5 TTS (~75ms). Built multi-room WebSocket architecture (host/speaker/listener roles) with API key pool rotation, real-time cost tracking, and persistent latency logging. [Demo]
View ProjectLLM SEO AI-Native Content Engine for Citation Optimization
January 1, 2026 – Present
Architected multi-stage AI content engine that researches, verifies, and generates citation-optimized articles directly cited by ChatGPT, Claude, Gemini, and Perplexity; engineered 5-stage pipeline: Question Discovery → Source Authority Mapping → Fact Verification → Hub & Spoke Knowledge Map → Article Generation. Implemented hallucination-prevention layers: quote-traceable fact extraction, cross-article contradiction detection, citation-to-content validation, and resume-from-checkpoint for cost-safety. [Demo]
View ProjectOffline LLM on Android Edge AI Inference
January 1, 2026 – Present
Engineered on-device LLM inference system deploying LFM 2.5 1.2B on Android (Poco X3) via llama.cpp + CMake — fully offline inference with zero internet dependency, running entirely on consumer mobile hardware. Proved edge AI viability: quantized LLM runs on-device with no cloud backend; tested and demoed to validate LFM 2.5 support on resource-constrained edge devices. [Demo] [Blog]
View ProjectAI Hoax Buster Chrome Extension
January 1, 2025 – Present
Built browser-integrated NLP Chrome extension for real-time bias and hoax detection with sub-800ms latency on news articles and web content. Engineered deterministic inference pipelines with chunked processing, label normalization, reproducible scoring, and manifest-compliant Chrome extension logic.
Introduction to Networks
Cisco
June 1, 2026 – Present
Data Analytics Job Simulation
Deloitte
June 1, 2026 – Present
Automation Developer
UiPath
June 1, 2026 – Present
IBM Data Science
Coursera
June 1, 2026 – Present
Google Python Crash Course
Coursera
June 1, 2026 – Present
AI Primer & Generative AI
Infosys
June 1, 2026 – Present
Cultural Fit Analysis
The candidate's diverse range of personal projects and internships, coupled with their active participation in publishing technical blogs, indicates a strong passion for AI/ML and a proactive, self-driven learning style. Their focus on open-source AI development and bridging research with practical engineering aligns well with an innovative and impact-driven culture. The variety of roles and technologies explored suggests a broad interest and willingness to tackle different challenges, contributing positively to team diversity and knowledge sharing.
Soft Skills & Operational Fit
The candidate demonstrates strong problem-solving skills through complex project architectures (e.g., multi-stage content engine, real-time translation system). Their ability to benchmark, optimize, and reduce latency indicates a results-oriented and efficient approach. The detailed project descriptions and blog posts suggest good technical communication and documentation skills. The breadth of projects and internships indicates adaptability and a proactive learning attitude.