Guneesh Vats

AI Researcher

Hyderabad, Telangana, India9 mos experience

Key Highlights

Expert in building scalable ML systems across diverse domains.
Ranked 7th globally in IEEE ICASSP Grand Challenge 2025.
Strong research background with multiple IEEE publications.

Stackforce AI infers this person is a Machine Learning Engineer specializing in AI-driven solutions across various domains.

Contact

guneesh.vats@servicenow.com LinkedIn

Skills

Core Skills

Natural Language Processing (nlp)Applied Machine LearningComputer VisionMlopsData EngineeringAutomatic Speech Recognition SystemAudio Machine LearningReal-time Inference SystemsWeb Application Development

Other Skills

Amazon EC2Amazon Web Services (AWS)Authentication ProtocolsBatch ProcessingCI/CD for Machine LearningCUDAContent Management Systems (CMS)Continuous Integration and Continuous Delivery (CI/CD)Emotion Perception AnalysisFAISSFastAPIFine TuningGNNGaussian Mixture Model (GMM)Gitlab

About

I’m a Machine Learning Engineer with experience building applied ML and GenAI systems across NLP, speech, computer vision, and large-scale data pipelines. My work sits at the intersection of research and production, translating ideas into reliable, scalable systems. At ServiceNow and early-stage startups, I’ve worked on problems ranging from LLM-powered information extraction and RAG systems to speech recognition, multimodal evaluation pipelines, and high-fidelity data extraction. I enjoy owning ML systems end-to-end, from modeling and experimentation to deployment, evaluation, and MLOps. I also have a strong research background, with publications in IEEE venues, including work on change-point detection in coded communication systems, speech-based emotion analysis for mental health, and computer vision pipelines for sports analytics. I was ranked 7th globally in the IEEE ICASSP Grand Challenge 2025 and have been a Dean’s List awardee at IIIT Hyderabad. I’m motivated by solving hard problems and building robust, high-impact ML systems that work in the real world. Always excited to collaborate on meaningful machine learning and AI projects.

Experience

9 mos

Total Experience

4 mos

Average Tenure

Current Experience

Servicenow

Research Scientist

Jan 2026 – Present · 5 mos · Hyderabad, Telangana, India · On-site

Coschool

Machine Learning Engineer

Jul 2025 – Dec 2025 · 5 mos · Hyderabad · On-site

Developed a fast context-refresh detection module for conversational AI using cosine similarity and heuristic gating, improving baseline F1 by 10% (to 0.86) while maintaining millisecond(ms) level inference latency.
Built a Gaussian Mixture Model (GMM) based classifier to detect on-topic vs off-topic queries and assign relevant topic tags to user queries in chatbot conversations, achieving over 95% accuracy.
Designed a multimodal handwritten-answer evaluation pipeline combining YOLO-based diagram detection, Mathpix OCR transcription, and rubric-aligned LLM scoring, reducing manual review time by 60% and LLM compute costs by 40%.
Implemented an automated PDF-assessment workflow to correctly map multi-question handwritten submissions to their corresponding questions using layout heuristics and embedding-based matching, significantly reducing mismatch errors.

Computer VisionLarge Language Model Operations (LLMOps)Natural Language Processing (NLP)Applied Machine Learning

Alonzo ai

Founding Machine Learning Engineer

Jan 2025 – Jun 2025 · 5 mos · Hyderabad · On-site

Built a CI/CD pipeline for LLMOps to extract structured data from natural-language prompts and generate SQL/Mongo queries; reduced query latency by 40% by streamlining multi-stage parsing and backend execution with prompt-version control.
Created a sports CMS ETL pipeline using Google/ViT embeddings and LLMs to process legacy recordbooks, automating data extraction and reducing manual tagging workload by 78%.
Optimized a RAG pipeline using FAISS + SBERT for sports-stat retrieval, improving factual accuracy and lowering LLM API costs by 30%.
Developed a comprehensive LLM evaluation framework (92% extraction accuracy, 98% DB-write accuracy) using pytest, JMeter, and Locust for automated testing, stress evaluation, and reliability benchmarking.

MLOpsData EngineeringCI/CD for Machine LearningLLM Pipeline Automation

Adalat ai

Machine Learning Intern

Oct 2024 – Dec 2024 · 2 mos · New Delhi · Remote

Fine-tuned Whisper with LoRA-PEFT for noisy multi-speaker courtroom ASR; improved WER by 18% → 21.4% on a 30k-minute dataset through diarization-aware log-mel alignment.
Built a speaker-aware preprocessing pipeline with silence removal, speaker-turn segmentation, diarization, and structured JSON transcript generation.
Developed a real-time ASR inference backend using FastAPI and NVIDIA Riva; achieved 100ms latency via GPU-efficient decoding, log-mel caching, token prefetching, and optimized batch scheduling.

Automatic Speech Recognition SystemAudio Machine LearningReal-Time Inference SystemsMultispeaker Speech Processing

International institute of information technology hyderabad (iiith)

3 roles

Web Administrator

May 2024 – Jul 2024 · 2 mos · Hyderabad, Telangana, India · On-site

Managed and developed websites for the eMSIT program, a Coursera-offered MS Programme by IIIT Hyd for students all over the world. Implemented scalable architectures to support high user traffic.

Teaching Assistant

Jan 2024 – Apr 2024 · 3 mos · Hyderabad, Telangana, India · On-site

Teaching Assistant for the course - Software Systems Development

SQLWeb Application Development

Teaching Assistant

Aug 2023 – Dec 2023 · 4 mos · Hyderabad, Telangana, India · On-site

Teaching Assistant for the course Learning and Memory

Proxzar

Machine Learning Intern

Jan 2024 – Jun 2024 · 5 mos · Hyderabad · On-site

Built an end-to-end ontology-driven data extraction pipeline using fine-tuned LLaMA2 and domain-specific prompts to transform unstructured product text into schema-aligned JSON for 1M+ catalog items.
Fine-tuned LLaMA2, Mistral-7B, and OpenAI models using LoRA, synthetic product data, and ontology-aware context injection, improving schema adherence and reducing hallucinations in structured outputs.
Integrated extracted entities and relations into a Neo4j-based Knowledge Graph, leveraging NER, relation extraction, and graph embeddings to power semantic retrieval; improved search precision by 22%.