Muttalip sk

Muttalip sk

AI Engineer
India
English, Bengali
Full-time On-site

About Me

AI Engineer specializing in LLM-powered applications, RAG systems, multilingual NLP, and backend AI engineering.
Experienced in designing production-ready AI pipelines, secure data workflows, and scalable inference syste…

Experience

AI/ML Engineer

Nexdev IT Solution pvt · Kolkata , West Bengal , India
Apr 2026 - May 2026 · 30 days

<p>Developed a production-grade Vision-Language Model (VLM) pipeline to solve the real-world industry problem of manual data entry, automating layout analysis and data extraction for complex invoices and bank cheques.</p><p> Architected and set up the complete fine-tuning infrastructure for the Qwen model, optimizing it for highly accurate enterprise document parsing of Bengali and English text.</p><p> Engineered the data preparation workflow and generated high-quality synthetic data to train the model on diverse, real-world document layouts and code-mixed Bengali-English content.</p><p> Fine-tuned the Qwen multimodal architecture on custom datasets, achieving 98% accuracy on printed Bengali and English data and 96% accuracy on handwritten data.</p><p> Deployed the trained system to process production documents for Prime Bank and CarryBee, successfully transforming unstructured financial forms into reliable automated data.</p>

Associate AI Engineer (Intern)

RedPluto Analytics Private Limited · Pune , Maharashtra , India
Jun 2025 - Mar 2026 · 8 months

<p>Developed a production-grade AI pipeline for analysing multilingual customer support calls, automating transcription, entity extraction, and structured business analytics.</p><p> Designed a dual-stage LLM workflow using Gemini AI: (1) high-accuracy speech-to-text transcription and (2) extraction of financial trade activity, sentiment and business opportunity signals.</p><p> Built ETL workflows to store transcripts, metadata and ground-truth audit records in PostgreSQL, enabling quality verification and reporting.</p><p> Implemented rule-based and statistical matching logic to detect mismatches across price, quantity and execution timing.</p><p> Engineered an asynchronous parallel task execution pipeline with rate-limiting, reducing overall processing latency by approximately 60%.</p>

AI Research Engineer

Odia Gen AI · Pune , Maharashtra , India
Feb 2024 - Jun 2025 · 1 year 4 months

<p>Fine-tuned LLaMA3-based Large Language Models for Odia language understanding and generation, improving performance across downstream NLP tasks.</p><p> Built a custom Odia tokenizer, enabling better token efficiency and multilingual model compatibility.</p><p> Researched model compression and quantization approaches, achieving approximately 60% reduction in model size while maintaining task accuracy.</p><p> Developed training and evaluation workflows using Hugging Face Transformers and LangChain.</p><p> Contributing to the development of regional-language AI ecosystems supporting education, communication, and accessibility.</p>

Skills

Python PyTorch TensorFlow Scikit-learn Keras LangChain LangGraph RAG Systems LLM Evaluation Multi-Agent Systems LLaMA Fine-tuning Pre-training Llama-Factory NumPy Pandas Matplotlib Seaborn FastAPI Django Flask PostgreSQL ChromaDB Pinecone AWS Docker Git GitHub Linear Algebra Probability & Statistics Calculus Gemini API RDKit DeepPurpose ChemBERTa Streamlit SQLite ArangoDB AQL MySQL Hugging Face Transformers
Report this Profile?