Agentic AI & LLM Engineer in Hyderabad

Most AI work stalls between a promising demo and a system people rely on every day. My job is closing that gap. At Eject Solutions I designed an agentic AI architecture that the company adopted as its production blueprint for new builds, and I lead the team that ships on it.

Five autonomous agent pipelines now run in production through Autonomous Agency Workflows, recovering more than 20 hours of staff time every week. They take over repetitive operational work so people spend their time on judgement, not copy-paste.

• Architecture that gets reused. The agentic design became the template for new builds, not a one-off prototype.
• Faster delivery. I made model and pipeline deployment 40% faster and reduced sprint delays by 35%.
• Delivered ahead of plan. I coordinated a 6-person team to deliver a market-ready AI agent 80% faster than planned.

I build LLM-powered automation pipelines that replace repetitive manual workflows across client operations. Using LangChain and the Claude API, I automated content creation, client reporting and scheduling, saving 20+ hours a week and cutting client report delivery from 3 days to 4 hours.

The pattern I follow is simple: the model drafts from verified data, and a person reviews before anything goes out. That keeps automation fast without letting it publish numbers nobody checked.

Engineering and research feed each other in my work. Each project below has its own page with the problem, approach, stack and results.

• GBNI (Generative Brand Narrative Intelligence). My state-aware multimodal agent framework for emotionally adaptive brand communication, with two papers on SSRN (v1 and the v2 extension) and its production companion, Autonomous Agency Workflows.
• MITHRA. A self-directed R&D platform for responding to cybercrime quickly and turning incident signals into usable intelligence, built with Claude Code- and Codex-assisted engineering. Its manuscript is in preparation for a Scopus-indexed journal.
• Diabetes Risk Prediction web app. A peer-reviewed model (IJSRED, 2022) with 82%+ accuracy and 0.87 ROC-AUC, using a Random Forest tuned with GridSearchCV and SHAP explanations for every prediction, deployed as a Streamlit app.
• Lead generation data pipeline. A Python and SQL pipeline that extracts, cleans, de-duplicates and scores leads from four channels; it produced 3,000+ qualified leads and facilitated ₹1.4Cr+ in revenue.

• GBNI project
• MITHRA project
• Diabetes Risk Prediction project
• How the lead pipeline works
• GBNI explained
• Explaining every diabetes prediction

I work in Python and SQL for data and modelling, LangChain and the Claude API for LLM automation, scikit-learn and SHAP for explainable machine learning, Streamlit for interactive apps, and Power BI for the dashboards leadership actually reads. My Power BI work has tracked 15+ KPIs that informed 8+ leadership decisions.

I also use Claude Code- and Codex-assisted development workflows to build faster without lowering the bar on review.

• All projects
• Full bio

I sit where engineering meets the boardroom. I build ML, NLP and agentic LLM systems end to end, then work alongside C-suite leaders to decide where those systems should point next. That means starting from the business metric, not the model, and measuring the result in the terms leadership already reports.