Sanskruti Ise

AI Researcher

Pune, Maharashtra, India4 yrs 7 mos experience

Most Likely To SwitchHighly Stable

Key Highlights

Expert in multilingual AI systems and language models.
Proven track record in AI project leadership and mentorship.
Innovative solutions for real-world AI challenges.

Stackforce AI infers this person is a multilingual AI specialist with a focus on NLP and machine learning applications.

Contact

Skills

Core Skills

Large Language Models (llm)Natural Language Processing (nlp)Data ScienceMachine LearningWeb Development

Other Skills

Adobe PhotoshopData AnalysisFine TuningGenerative AINatural LanguagePhotographyProject ManagementPython (Programming Language)Retrieval-Augmented Generation (RAG)

About

I’m an AI Engineer with 4+ years of experience helping machines talk, listen, and occasionally behave themselves. I’ve worked on projects ranging from multilingual language models and speech systems to intelligent agents and AI platforms that actually solve real-world problems. Along the way, I mentor teams, tackle tricky data challenges, and turn research into practical, scalable solutions. I like my AI impactful, human-centered, and just a little clever, because why should machines have all the fun?

Experience

Dheeyantra

3 roles

Lead Research Engineer

Promoted

Dec 2024 – Present · 1 yr 3 mos

Built an internal 𝐀𝐈 𝐩𝐥𝐚𝐭𝐟𝐨𝐫𝐦 with RAG-powered knowledge management, a no-code agentic workflow builder, multi-voice TTS for virtual agents, and production deployment/testing tools.
Developed 𝐃𝐡𝐞𝐞𝐍𝐱𝐭𝐆𝐞𝐧 𝐋𝐋𝐌 – a trilingual (Hindi, English, Tamil) model trained on 50B tokens using H100s on Red Hat OpenShift, achieving ~1.4 loss.
Open-sourced 𝐒𝐅𝐓-𝐭𝐫𝐚𝐢𝐧𝐞𝐝 𝐌𝐢𝐬𝐭𝐫𝐚𝐥-𝟕𝐁 𝐰𝐞𝐢𝐠𝐡𝐭𝐬 in Hindi, Marathi, Tamil, and Malayalam using Unsloth.
Created an 𝐚𝐠𝐞𝐧𝐭𝐢𝐜 𝐋𝐋𝐌 𝐬𝐲𝐬𝐭𝐞𝐦 for flight bookings with Mistral-7B, quantized to 4 bits.
Generated 𝐬𝐲𝐧𝐭𝐡𝐞𝐭𝐢𝐜 𝐝𝐚𝐭𝐚 for Hindi & Tamil using Nvidia Curator.
Trained a new 𝐆𝐏𝐓-𝟐 𝐭𝐨𝐤𝐞𝐧𝐢𝐳𝐞𝐫 on OpenWebText + Sangraha (Hindi/Tamil) corpus.
Co-authored a 𝐫𝐞𝐬𝐞𝐚𝐫𝐜𝐡 𝐩𝐚𝐩𝐞𝐫 on transformers for chronic pain management in cancer patients (submitted to NSF with MY Jeeva Care).
𝐋𝐞𝐝 & 𝐦𝐞𝐧𝐭𝐨𝐫𝐞𝐝 a team of 8 engineers and interns.

Retrieval-Augmented Generation (RAG)Large Language Models (LLM)Natural Language Processing (NLP)Machine LearningData SciencePython (Programming Language)

Senior Research And Development Engineer

Promoted

Nov 2022 – Dec 2024 · 2 yrs 1 mo

Trained 𝐎𝐩𝐞𝐧𝐀𝐈 𝐖𝐡𝐢𝐬𝐩𝐞𝐫 for 8 Indic languages, achieving WER < 15 on Common Voice.
Fine-tuned 𝐎𝐏𝐓-𝟏.𝟑𝐁 (loss 1.5) and 𝐆𝐏𝐓-𝐍𝐞𝐨 (loss 2.3).
Achieved 𝟕𝟎% faster training for RoBERTa with distributed training (4 nodes, 2 GPUs each) while improving emotion classification accuracy to 96%.
Built an 𝐢𝐧𝐭𝐞𝐫𝐚𝐜𝐭𝐢𝐯𝐞 𝐰𝐞𝐛 𝐤𝐢𝐨𝐬𝐤 𝐚𝐯𝐚𝐭𝐚𝐫 using Character Creator & Blender.
Migrated 𝐓𝐚𝐜𝐨𝐭𝐫𝐨𝐧 → 𝐅𝐚𝐬𝐭𝐒𝐩𝐞𝐞𝐜𝐡 pipelines for multilingual TTS; contributed Oriya voice recipes in ESPnet.
Engineered a 𝐰𝐚𝐤𝐞 𝐰𝐨𝐫𝐝 𝐝𝐞𝐭𝐞𝐜𝐭𝐢𝐨𝐧 𝐦𝐨𝐝𝐞𝐥 (“Hey Dhee”, “Dheeyantra”) with 90% accuracy.
Curated a 𝐩𝐚𝐢𝐧 𝐝𝐞𝐭𝐞𝐜𝐭𝐢𝐨𝐧 𝐝𝐚𝐭𝐚𝐬𝐞𝐭 and fine-tuned VGG19 for facial expression–based severity classification.

Fine TuningMachine LearningNatural Language Processing (NLP)Data SciencePython (Programming Language)

Research And Development Engineer

Aug 2021 – Nov 2022 · 1 yr 3 mos

Deployed a 𝐜𝐮𝐬𝐭𝐨𝐦 𝐞𝐦𝐨𝐭𝐢𝐨𝐧 𝐫𝐞𝐜𝐨𝐠𝐧𝐢𝐭𝐢𝐨𝐧 𝐦𝐨𝐝𝐞𝐥 (73% accuracy overall; >90% top-5).
Implemented 𝐨𝐩𝐭𝐢𝐜𝐚𝐥 𝐟𝐥𝐨𝐰 techniques for behavior tracking & report generation.
Built 𝐂𝐑𝐅-𝐛𝐚𝐬𝐞𝐝 𝐍𝐄𝐑 and 𝐇𝐌𝐌-𝐛𝐚𝐬𝐞𝐝 𝐏𝐎𝐒 tagging models for Indic languages with >90% accuracy.
Designed and deployed a 𝐝𝐚𝐭𝐚 𝐜𝐨𝐥𝐥𝐞𝐜𝐭𝐢𝐨𝐧 𝐩𝐢𝐩𝐞𝐥𝐢𝐧𝐞 for multimodal AI research.

Natural Language Processing (NLP)Machine LearningData Science