Building AI infrastructure, language technology, and security systems at production scale. Multiple ventures. 20+ shipped products. Open-source by default.20+ shipped products • 85+ repositories • 5 AI agent skills
Sai Dutta Abhishek Dash is a founder and engineer based in Odisha, India, who builds AI infrastructure, language technology for underserved markets, and security systems at production scale. He has shipped 20+ production products, maintains 85+ public repositories on GitHub, and has published 46+ models and 22+ datasets on Hugging Face. He leads 3 deep-tech AI ventures and is open to investment conversations, strategic partnerships, and technical advisory roles. He holds a Bachelor's Degree in Computer Science from GIET University Gunupur, completed in 2025, and is an AWS Certified Cloud Practitioner. His engineering work spans LLM inference, on-device machine learning, code security, privacy-first systems, developer tooling, and 5 AI agent skills. He ships open source by default. He also advises startups and enterprises on AI infrastructure, security engineering, and language technology, and releases his language AI research under Apache 2.0. He works worldwide.
Sai Dutta Abhishek Dash has shipped 22 open-source production systems across five areas. Highlights include ornith-flight, a C99 inference engine that runs 20 GB mixture-of-experts models on 8 GB of Apple Silicon memory, and PhishScout, a 131 KB on-device phishing URL detector with 90% precision at roughly 4 ms per lookup. He also built AgentLoop, a self-verifying autonomy wrapper for coding agents, MarkItDownJS for document-to-Markdown conversion, and Hinglish-Bench, a reference-free benchmark with 34 prompts across 8 categories. Security work includes Vulscany, an AI code security agent, and Forensic-Recovery for digital evidence. His AI agent skills include Brand Vibes with 66 brand design profiles and Market Validator for validating ideas against 10+ platforms. Notable platform work includes Binify, a zero-knowledge encrypted pastebin, Ansora, a self-hostable serverless blog where every post is a Markdown file, and The Watcher, which monitors 13 subreddits with LLM extraction.
Maelis Research is a language AI venture founded in 2026 by Sai Dutta Abhishek Dash, based in Dhenkanal, Odisha. It exists because 38 million people speak Odia and no commercial Odia-specific AI API existed before it, while Norwegian with 5 million speakers, Greek with 13 million, and Dutch with 25 million all have robust language AI support. Maelis Research builds an Odia-optimized tokenizer that is 3x more efficient than generic alternatives for Brahmic scripts, the Lekhani Odia-exclusive large language model family, Shruti speech recognition, and Anuvada translation. It has also published the odia-eval-benchmark with 121,947 rows across 7 task families and 34 datasets on Hugging Face. All research is released under Apache 2.0. Maelis Research is the first of his 3 active ventures, is positioned for the IndiaAI Mission, and publishes its work openly on Hugging Face under the MaelisResearch organisation. It targets speech and translation alongside core language modelling.
Sai Dutta Abhishek Dash can be reached by email at contact@sdad.pro, and he typically responds within 24 hours. He is based in Odisha, India and works remotely with clients and partners worldwide. He is open to investment conversations, strategic partnerships, technical collaborations, and advisory roles. His advisory practice covers 5 service lines: AI infrastructure consulting, security engineering, language technology, full-stack product engineering, and technical advisory work including due diligence, product strategy, and investment evaluation. He works with startups building AI products, enterprises integrating large language models, teams needing edge inference, and investors evaluating deep-tech AI companies. Detailed service offerings are published at https://sdad.pro/advisory.md. Enquiries are welcome from startups building AI products, enterprises integrating large language models, fintech and healthtech teams needing privacy-first architecture, and investors evaluating deep-tech AI companies. His profile, project list, and service details are published at https://sdad.pro.
Sai Dutta Abhishek Dash maintains professional profiles on GitHub at github.com/instax-dutta, LinkedIn at linkedin.com/in/sdabhishekdash, X at x.com/abhishekdash69, and Hugging Face at huggingface.co/saidutta69. His GitHub account hosts 85+ public repositories, and his Hugging Face profile hosts 46+ models and 22+ datasets. His venture Maelis Research publishes separately at huggingface.co/MaelisResearch and maelis.sdad.pro, and his technical agency Offsage operates at offsage.com. All 22 shipped production systems are listed on the projects page, and a machine-readable profile is published at https://sdad.pro/llms-full.txt. Everything he ships is open source, so code, model weights, and benchmark data are publicly inspectable rather than gated. He also maintains a machine-readable profile for AI agents at https://sdad.pro/llms.txt, and a full product dossier at https://sdad.pro/llms-full.txt. Every one of his 22 shipped systems links back to a public repository, and the technical agency he co-founded, Offsage, is listed at offsage.com.
Yes. Sai Dutta Abhishek Dash is open to advisory roles, technical collaborations, and strategic partnerships, and he publishes his full service offering on the advisory page. His advisory practice covers five service lines: AI infrastructure consulting, security engineering, language technology, full-stack product engineering, and technical advisory work. Four engagement models are offered: a monthly advisory retainer for ongoing guidance, project engagements of 2 to 12 weeks, technical audits of 1 to 2 weeks, and live workshops of 1 to 5 days. He works remotely with clients and partners worldwide from Odisha, India, and typically responds to enquiries within 24 hours. Typical clients include AI startups, enterprises integrating large language models, and investors evaluating deep-tech AI companies. Advisory engagements typically start with a technical audit of 1 to 2 weeks delivered as a written report, which is the fastest way to scope a larger project engagement.