I build production-oriented AI systems across multi-agent orchestration, RAG, LLM fine-tuning, Production ML, and streaming inference. My strongest work combines research discipline with deployable software.
The strongest projects lead with the problem, the engineering mechanism, and the proof. This is the order recruiters, founders, and research teams need.
A procedural SRE training environment for cooperative LLM agents handling cascading microservice failures.
Emergency triage prediction system with hierarchical safety logic and clinical feature engineering.
Limit order book prediction system rebuilt after diagnosing batch-to-live degradation.
AI-native civic waste reporting platform with vision, fallback routing, voice input, and city-level workflows.
Non-partisan election assistant for Indian voters with multilingual guidance and ECI-aligned fact checks.
Agentic code analysis tool with explain, test, and fix modes for developer workflows.
The important story is not one lucky ranking. It is repeated performance across prompt engineering, agents, clinical ML, and quantitative streaming inference.
Built a grounded support triage agent using only the provided corpus, with evidence retrieval and auditable outputs.
National prompt engineering and AI system design competition at REVA University, Bengaluru.
Quantitative AI challenge focused on limit order book prediction under live streaming constraints.
Built under hackathon constraints with focus on useful AI product execution.
Clinical ML challenge with hierarchical modeling and high-stakes triage metrics.
The stack is grouped by the problems it solves. This reads stronger than a flat wall of tools.
Agent orchestration, retrieval, tool use, prompt contracts, evaluation, grounding, and model fallback.
Clinical ML, time-series modeling, self-supervised learning, fine-tuning, ranking metrics, and ablations.
APIs, frontends, deployment, integrations, streaming inference, and pragmatic reliability tradeoffs.
These are useful secondary signals, but the primary narrative remains production AI systems.
Phi-2 QLoRA, NIFTY options data, deterministic orchestration, walk-forward evaluation, and RAG ablation.
View repoCPU vs CUDA Dynamic Time Warping on Bitcoin time-series data, showing low-level parallelization depth.
View repoSatirical AI evaluation product with structured scoring and sharp product writing. Useful personality signal.
View liveI am a Fourth-year B.Tech CSE student at IIIT Nagpur. My work sits at the intersection of agentic AI, RAG, reinforcement learning, Production ML, and production engineering.
The pattern across my projects is deliberate: build a system, expose the failure modes, measure behavior, and deploy a usable version. That is why my strongest work includes live systems, competition rankings, documented model choices, and clear evaluation metrics.
I am looking for AI engineering internships, applied research roles, and founder-led teams where the work demands both technical depth and product judgment.