Headline

Building Multilingual Voice AI & Agentic Systems | Ex-AI Engineer Intern @Aumin Labs | Ex-AI Research Intern @Teesside University | LLMs & Agentic AI | Microsoft Certified: Azure AI Fundamentals

About

💻 𝐀𝐈/𝐌𝐋 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫 | 𝐒𝐩𝐞𝐞𝐜𝐡 & 𝐍𝐋𝐏 | 𝐁𝐮𝐢𝐥𝐝𝐢𝐧𝐠 𝐒𝐜𝐚𝐥𝐚𝐛𝐥𝐞, 𝐑𝐞𝐬𝐩𝐨𝐧𝐬𝐢𝐛𝐥𝐞 𝐀𝐈 𝐒𝐲𝐬𝐭𝐞𝐦𝐬 I'm driven by how AI can turn noisy, real-world data into reliable systems — from multimodal biomedical research to production speech pipelines. My focus is always on practical impact, interpretability, and reproducibility. 🔬 𝐄𝐱𝐩𝐞𝐫𝐢𝐞𝐧𝐜𝐞 & 𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬 Aumin Labs LLP — ML & Speech Engineer Intern building NeoVani, a real-time speech-to-speech translation system for Indian languages (ASR → NMT → TTS). Teesside University — Built a multimodal framework fusing spatial genomics with histopathological imaging for breast-cancer insights. Open Health Systems Laboratory — Led NLP pipelines for digitizing and structuring ancient Ayurvedic texts. Digital Hercules Innovations — Fine-tuned BERT-based keyword-extraction models, achieving a BLEU score of 28 for digital-marketing applications. 🔧 𝐓𝐞𝐜𝐡𝐧𝐢𝐜𝐚𝐥 𝐒𝐤𝐢𝐥𝐥𝐬 Speech & ASR — NeMo, FastConformer, CTC/RNNT architectures, streaming ASR, tokenizer design Core ML & Multimodal — PyTorch, TensorFlow, Hugging Face Transformers, vision + genomics + text fusion LLM/NLP Tooling — LangChain, prompt engineering, LoRA/adapters, IndicTrans2, CTranslate2 MLOps & Deployment — Docker, FastAPI, CI/CD, MLflow/W&B Data & Evaluation — Pandas/NumPy/SQL, SpaCy, BLEU, WER, SHAP/Captum 🚀 𝐏𝐫𝐨𝐣𝐞𝐜𝐭𝐬 & 𝐈𝐦𝐩𝐚𝐜𝐭 NeoVani: real-time speech-to-speech translation, ~1.2s end-to-end latency across 5 Indian languages Multimodal Breast-Cancer Insights: spatial-genomics + histopathology fusion Ayurvedic Text NLP: OCR → normalization → entity extraction Marketing Keyword Extraction: BERT fine-tuning, BLEU 28 📩 Open to research-adjacent ML/NLP roles — reach out via LinkedIn. Skills: Data Science | Speech & ASR | NLP | Deep Learning | Generative AI | Biomedical Analytics

Location

Uttar Pradesh

AI stack

anthropic apis

antigravity

claude code

hugging face

langchain

langgraph

Work experience

AI Engineer Intern at Aumin Labs, Jan 2026 - Jul 2026 · 7 mos, Bengaluru, - Engineered a real-time, multi-stage speech-to-speech translation framework for Indian languages, integrating automatic speech recognition, neural machine translation, and text-to-speech synthesis to address low-resource Indic language accessibility and cross-lingual communication challenges. - Fine-tuned an English-base ASR model (NVIDIA NeMo Conformer-CTC) through a reproducible, environment-variable-driven training pipeline, reaching 9.10% WER / 2.83% CER at a real-time factor of 0.025 (~39× real-time); designed and began migrating the ASR stage to a cache-aware streaming FastConformer (~114M parameters) with a 10k-unigram SentencePiece tokenizer, intermediate-CTC supervision, and frame-level language identification for low-latency on-device inference. - Benchmarked translation quality against baseline NMT systems fine-tuned on a 1.19M-pair parallel corpus, reaching BLEU scores up to 19.04, and validated end-to-end pipeline latency averaging ~1.2s across 2,464 samples for live conversational translation. - Ensured reliability and traceability through persistent subprocess workers over line-delimited JSON IPC with DuckDB session logging for per-stage latency tracing, and engineered a script-disambiguation layer to resolve Devanagari/Latin transliteration routing errors.

AI Research Intern at Teesside University, May 2025 - Jul 2025 · 3 mos, Middlesbrough, England, United Kingdom, - Engineered a multimodal, Retrieval-Augmented Generation (RAG) agentic framework for breast cancer insight generation, integrating spatial genomics, histopathological imaging, and biomedical literature to address key challenges in diagnostic interpretation and research fragmentation. - Developed and fine-tuned specialist AI agents using advanced contrastive learning techniques (SimCLR, SimCSE) on heterogeneous biomedical datasets—including 10X Genomics Visium data and 200,000+ PubMed cancer abstracts—resulting in high-performance retrieval and enhanced interpretability of complex clinical queries. - Benchmarked novel agent architectures against state-of-the-art biomedical AI models, achieving significant improvements in recall@k and interpretability with a structured, multi-agent pipeline designed for real-time cross-modal fusion of text and spatial omics data. - Ensured transparency and explainability through rigorous logging of agent inferences, facilitating traceable, clinically actionable insights and positioning the project as a pioneering agentic multimodal system in biomedical AI.

Summer Intern at Open Health Systems Laboratory(OHSL), May 2025 - Jul 2025 · 3 mos, Los Gatos, California, United States, - Led a team in developing and reviewing cutting-edge NLP pipelines for processing Sanskrit Ayurvedic texts, guiding the application of rule-based, statistical, and neural methods for tasks such as sandhi splitting, morphological analysis, named entity recognition, relation extraction, and machine translation within a low-resource domain. - Conducted an extensive, peer-reviewed survey spanning over 100 studies from 2000–2025, evaluating more than 25 tools and frameworks—including transformer models, BERT variants, and retrieval-augmented QA systems—using metrics like accuracy, F1, BLEU, and TER to benchmark performance on Sanskrit–Ayurveda corpora. - Spearheaded efforts to catalogue and compare novel NLP methods (e.g., mT5, ByT5, SansGPT, AyuRAG) targeting digitization and deep semantic understanding of classical medical texts, overcoming key challenges in OCR error correction, data scarcity, and cross-lingual translation. - Organized and delivered actionable recommendations for hybrid rule-neural architectures, data creation pipelines, community benchmarking initiatives, and protocols for model explainability, fostering collaborative workflows between computer scientists, Sanskrit experts, and clinicians. - Enabled practical deployment of AI tools to unlock millennia-old Ayurvedic wisdom, supporting advanced search, automated translation, and clinical decision support for researchers and practitioners; championed open science and robust documentation throughout project delivery and communication.

Data Science Intern at Kernel Synergy, Jan 2025 - Feb 2025 · 2 mos, Delhi, India, - Served as a Data Science Intern, addressing a range of real-world problem statements by designing and developing specialized libraries to enhance data science and machine learning workflows for advanced analytical solutions. - Led innovative AI/ML solution development, applying theoretical knowledge to practical challenges, and contributing to the advancement of in-house analytics capabilities. - Collaborated in a team-oriented environment, leveraging modern data science techniques to drive project success and deliver impactful results for diverse business needs. - Strengthened proficiency in Python, library development, and experimental design, building a solid foundation for ongoing growth in the field of artificial intelligence and data science.

AI Developer Intern at Digital Hercules Innovations, Aug 2024 - Dec 2024 · 5 mos, Noida, Uttar Pradesh, India, - Developed and fine-tuned BERT-based keyword extraction models using custom datasets, achieving a BLEU score of 28 and significantly improving the accuracy and precision of data extraction for digital marketing applications. - Designed and optimized machine learning pipelines, applying advanced feature engineering techniques to enhance model efficiency and predictive performance within real-world Amazon Ads optimization tasks. - Collaborated closely with cross-functional teams to implement cutting-edge AI methodologies, driving measurable improvements in digital marketing strategies and ad performance for major ecommerce campaigns.

Education

Bachelor of Technology - BTech, Computer Science & Engineering, amity University

12th Grade, PCM + Computer Science, ryan international school

Public links

github: https://github.com/ambermanijha