Lokesh Goenka


Lokesh Goenka

AI Engineer | Speech AI • Conversational AI • Generative AI

More About Me


I'm an AI Engineer focused on building intelligent systems across Speech AI, Conversational AI, and Generative AI.

Over the past year, I've built real-time conversational AI systems, voice agents, and LLM-powered applications, working with Speech-to-Text (STT), Text-to-Speech (TTS), Audio-to-Audio models, Retrieval-Augmented Generation (RAG), and frameworks such as Google ADK, LiveKit, and AWS.

More recently, my work has focused on TTS model fine-tuning and speech research, exploring speech representations, codecs, conditioning strategies, and model architectures for higher-quality, domain-specific speech generation.

Before moving into Speech AI, I worked with LLMs, RAG, and Educational AI. During my Master's in Data Science and Computing at SSSIHL, I developed a Personalized Learning Assistant for adaptive assessment and question generation using large language models.

I enjoy working at the intersection of research and engineering — understanding how models work, experimenting with them, and turning them into useful systems.


My professional work across LLMs, RAG, Conversational AI, Speech AI, and Generative AI.

Soulax

Associate AI & ML Engineer

August 2025 - Present

  • Build real-time conversational AI agents using Google ADK, LiveKit, and AWS, focusing on low-latency and context-aware voice interactions.
  • Evaluate and integrate Speech-to-Text (STT), Text-to-Speech (TTS), and Audio-to-Audio models for production voice applications.
  • Fine-tune and experiment with TTS models, including speech representations, codecs, conditioning strategies, and model architectures.
  • Develop agent workflows involving tool calling, contextual responses, and natural multi-turn conversations.
AI & ML Intern

May 2025 - July 2025

  • Developed LLM-powered NL2SQL pipelines for converting natural-language questions into executable SQL queries.
  • Contributed to Retrieval-Augmented Generation (RAG) pipelines for grounded and context-aware responses.
  • Researched reinforcement learning approaches for improving agent performance and user interactions.

SIFY

Data Science Intern

June 2024 - August 2024

  • Contributed to a personalized learning platform for adaptive student learning and automated assessment generation.
  • Fine-tuned FLAN-T5 and Llama 3.1 models for automated multiple-choice question generation.
  • Achieved 95% generation accuracy on educational assessment datasets.



A selection of research and engineering work.

Speech AI Research

TTS • Codec Models • LLM Fine-tuning

Vedic Sanskrit TTS

Research Project | 2026 - Present

Tech Stack: Python, PyTorch, LLMs, Speech AI

Researching and developing a TTS pipeline for Vedic Sanskrit mantra chanting, with emphasis on pronunciation, Vedic svara, and high-fidelity speech generation. Experimenting with codec-language-model architectures, Indic-Mio/MioCodec, LoRA fine-tuning, codec tuning, and speech evaluation workflows.


Personalized Learning Assistant

Personalized Learning Assistant for Students

Master's Project | Sri Sathya Sai Institute of Higher Learning (SSSIHL)

Tech Stack: Python, PyTorch, Hugging Face Transformers, FAISS, LLaMA, QLoRA

Developed an AI-powered Personalized Learning Assistant that identifies student strengths and weaknesses through adaptive assessments. Fine-tuned Llama 3.1 using QLoRA for MCQ generation, implemented RAG with FAISS for educational content retrieval, and explored reinforcement-learning strategies for personalized question generation.


Saavaad Conversational AI

Saavaad Conversational AI

Bachelor's Group Project | SSSIHL

Tech Stack: Python, PyTorch, Hugging Face Transformers, Google T5

Developed a website for translation, summarization, and text-to-speech, aimed at helping users overcome language barriers and access information more easily.


StorEtellor

StorEtellor | Moral Storytelling Platform for Children

Personal Project

Tech Stack: HTML, CSS, JavaScript

My first personal project, StorEtellor is a platform where children can read stories designed to communicate meaningful values and lessons.





Research from my work in Educational AI, personalized learning, and adaptive assessment.

Question Generation for Personalized Learning Platform for Students

Springer Nature | 2026

Research on automated question generation for personalized learning and adaptive student assessment.

From Uniform to Unique: Adaptive K–12 Assessment Using Large Language Models

Lecture Notes in Computer Science (LNCS) | 2025

Research exploring large language models for adaptive K–12 assessment and personalized learning experiences.


Some of the work and interests that matter to me beyond my technical career.

Eco-Einsteins

Co-Founder

February 2023

  • Co-founded a student-led initiative focused on environmental sustainability and responsible waste management.
  • Organized events and workshops to promote sustainable living and environmental awareness.
  • Worked with students to encourage practical, everyday approaches to reducing waste and building eco-conscious habits.

Connect with me!



Socials