Headline

AIML Consultant at South Asian Institute of Crime and Justice Studies

About

Currently contributing to SAICJS's Suraksha Lens by providing consultant and building and equipping an already existing dashboard with 7 regional languages making it more accessible for the people on ground to use our dashboard. This makes the dashboard more welcoming in and for the most rural public and areas of South Asia( Southern India and Sri Lanka) where the hint of modern language hasn't left it's mark yet. I used json files and i18n python library to achieve and cater to our major concerns: 1. the dashboard should be available in the slowest and rural most regions where the internet amounts to luxury that is not present every time and; 2. it should increase inclusivity of the people for whom this whole things is being run for. This solution, does not add any weight to the dashboard and every translation is being handled at the lowest level, which seemed right fit for our use case and also seemed most efficient. As per the promised made to ourselves before making it to the UNICEF, and making our dashboard AI powered, I have recently led the development of the chatbot that, for level 1, guides the new users and old through the dashboard and helps them understand the data, textual and graphical. -> I made the choice of maintaining screenshots of the current state of the dashboard which would contain everything on the screen, including country/state filters and indicators. -> Curated system prompt a 1000 times to make sure our model: openai/gpt-oss-120b - on_demand is token efficient even with all the graphical data( which is a lot when it is communicated to LLM). -> To monitor change in state, current screen state is compared with a previous screen state, it lead to an edge case where the model went over token limit as calculating the diff in the screen state would duplicate the entire chat_data but we got through it. This is the kind of work that I enjoy doing, this is the kind of work that I look forward to doing and if anyone is looking for someone for this kind of work or are open to build onto anything interesting in this space and looking for someone to join them or just talk, I'm a DM away.

Location

India

AI stack

claude

hugging face

langchain

mcp

Pinecone

PydanticAI

Work experience

AIML Consultant at South Asian Institute of Crime and Justice Studies, Jul 2026 - Present · 2 mo, Remote, Architected and deployed a production multilingual analytics dashboard supporting 7 regional languages using a scalable i18n system (JSON-based locale architecture, dynamic language-switching, automated translation validation pipeline), expanding platform accessibility for a national base of researchers, law enforcement stakeholders, and policymakers. Designed and integrated an LLM-powered conversational chat-bot enabling natural-language Q/A grounded in both dashboard text content and data visualizations, reducing average analyst time-to-insight by ~35% and removing the need for manual chart interpretation by non-technical users. Built the chatbot's context-grounding and retrieval pipeline (FastAPI backend, structured prompt engineering, Pydantic-validated outputs) to deliver accurate, context-aware responses across all 7 supported languages while maintaining sub-2-second response latency at scale.

Artificial Intelligence Technologist at BackkBenchers, Jan 2026 - Mar 2026 · 3 mos, Remote, • Key Achievement:Produced 60+ AI-generated video ads across short and long-form formats for 3 brands, leveraging Gemini for multimodal storyline generation and Veo 3.0 for prompt-to-video synthesis at production scale, sustaining a delivery rate of 2–3 production-ready assets daily. • Owned creative pipeline from client brief interpretation to prompt engineering, asset generation, and iteration delivering short-form and long-form ad creatives that achieved 3K–17K organic views per post on Instagram. • Drove multimodal content production at agency scale by architecting prompt workflows optimized for brand-aligned visual storytelling, maintaining near-zero revision cycles through precise prompt engineering and creative alignment with client vision.

AI Assistant Research Engineer at Ekanga, Sep 2025 - Dec 2025 · 4 mos, Gwalior, Madhya Pradesh, India

Artificial Intelligence Engineer at Ekanga, Sep 2025 - Dec 2025 · 4 mos, • Key Achievement: Collaborated directly with the CEO to architect a multi-modal AI Avatar system, integrating lifelike image and video generation pipelines. • Engineered 500+ hyper-realistic prompts across Stable Diffusion, Midjourney, and Gemini; iterative fine-tuning improved visual fidelity and thematic alignment by 40% over baseline. • Built a systematic prompt refinement pipeline cutting inconsistencies by 30% and iteration time by 25%; led video generation experimentation on Heygen.ai and Veo3.

Research Assistant at Ekanga, Sep 2025 - Dec 2025 · 4 mos, • Collaborated directly with the CEO to develop an AI Avatar, leveraging multi-modal AI systems for lifelike image and video generation. • Designed and refined over 500+ high-quality prompts to achieve hyper-realistic image outputs using Stable Diffusion, Midjourney, Gemini, and other AI image-generation LLMs. • Conducted iterative fine-tuning and comparative evaluation of model outputs to improve visual fidelity and thematic alignment by 40% over baseline generations. • Led experimentation with video generation platforms such as Heygen.ai and Veo3, optimizing speech delivery, facial expressions, and realism in lip-sync performance. • Developed a systematic prompt refinement pipeline, reducing generation inconsistencies by 30% and cutting iteration time by 25%.

Web Development Intern at Persist Ventures, Mar 2024 - Sep 2024 · 7 mos, • Collaborated closely with clients and a small development team to develop custom websites to fulfill client needs. • Performed QA and bugfixes on the team’s custom web application ensuring a seamless user experience. • Stack: JavaScript, React, TailwindCSS.

Back End Developer at discourse.work, Sep 2023 - Mar 2024 · 7 mos, Bengaluru, Karnataka, India

Education

Bachelor of Technology - BTech, Cyber Security, ITM University, Gwalior

Public links

adarshportfolio-xi.vercel.app: https://adarshportfolio-xi.vercel.app/