Building production AI systems that reason, retrieve, and act — not just autocomplete.
I'm a final-year CSE student at SRM Institute of Science and Technology (CGPA 9.39/10, graduating May 2027) who ships AI systems end-to-end — agent orchestration, RAG pipelines, evaluation infrastructure, and the production APIs underneath all of it.
My focus is agentic systems that have to be trusted, not just impressive — which means the evaluation and failure-handling layer matters to me as much as the agent logic itself. I've spent the last year doing this in production: rebuilding a brittle SQL agent into a supervisor-managed, async-streaming LangGraph pipeline, and running adversarial testing on production LLM agents to surface failure modes before they reached clients.
AI Engineer Intern · RMgX Technologies — May 2026 – Jul 2026 Built a domain-agnostic conversational chatbot end-to-end — owned recognition/understanding, response generation, and integration, designed to generalize across use cases rather than a single hardcoded flow. Alongside this, worked on a document/data ingestion pipeline handling messy real-world edge cases, integrating with existing infra and Table Transformer (TATR) for structured extraction, with hands-on experimentation using YOLO for detection tasks.
AI Engineer Intern · RMgX Technologies — Jun 2025 – Jul 2025 Ran adversarial testing on production LLM agents — engineering edge-case prompts that broke SQL agent query logic and induced hallucinations, surfacing failure modes prior to client deployment. Rebuilt a brittle SQL agent into a supervisor-managed, async-streaming LangGraph agent, eliminating recurring production failures. Shipped 3 production-ready agentic components (RAG, SQL, and internet agents) across 4 sprints in a 5-person cross-functional team.
| Project | What it does | Stack | Key Result |
|---|---|---|---|
| Aletheia · Live Demo | Self-hosted GraphRAG system reconstructing company events across Slack, Jira & GitHub — two-pass linking (deterministic + IDF-weighted Jaccard + optional LLM pass) with auditable confidence scores on every edge | FastAPI · Neo4j · Qdrant · PostgreSQL · Redis · LangGraph · Gemini | Hybrid retrieval (BM25 + dense RRF → graph expansion → reranking), RBAC-gated end to end |
| CMPT · Live Demo | AI competitive pricing platform — deterministic pricing engine handles rule-based decisions, an Ambiguity Agent resolves edge cases, an Explanation Agent generates auditable rationale | CrewAI · LangGraph · Gemini · Firecrawl · FastAPI · PostgreSQL | ~90% reduction in manual competitor data collection |
| Dumb It Down · Live Demo | Hybrid RAG pipeline over a ~1,250-paper PubMed corpus, converting dense academic text into accessible summaries | FAISS · Qdrant · Gemini 2.0 Flash · FastAPI · React · SQLite | 3-stage retrieval (rewrite → search → rerank), validated via user testing |
LLM & Agents
Retrieval & Evaluation
Backend & Infra
Frontend
- 🟠 AWS Certified AI Practitioner (2025)
- ☁️ AWS Certified Cloud Practitioner (2025)
📄 CNN & DNN for Alzheimer's Detection via EEG — Independent research, joint first author. State-of-the-art 69.22% accuracy on BCI Competition III Dataset IV (subject-independent LOSO evaluation); 86.89% under subject-dependent evaluation.
- 🤝 Open to AI/ML engineering internships — agentic systems, RAG, LLM evaluation. Remote or India-based.
"The best way to understand a system is to build one."