I am a Computer Science student at IIIT Nagpur and currently an AI Engineer Intern at Neurema. I build practical AI systems, focusing on real-world problems with measurable results, from real-time inference and agentic pipelines to ranking systems and public-data analytics.
I currently work on projects that combine:
- Computer Vision and real-time inference
- Agentic RAG, semantic search, and multilingual assistants
- Recommendation and ranking systems
- Cloud backend engineering, APIs, and scalable deployment
WhatsApp • LangGraph • Gemini • Qdrant • Serper API • Render
- Built multilingual, voice-enabled agentic assistant handling 22 Indian languages .
- Implemented a LangGraph reasoning agent to dynamically chain semantic search, web search, and eligibility tools .
- Achieved ~30s latency for 20s audio inputs via optimized STT and retrieval pipelines .
Agent Orchestration • Self-Healing Validation • AWS EC2 • Bedrock
- Architected a 6-layer autonomous pipeline generating math-backed Manim videos under ~3 minutes .
- Executed AST static checks and LLM-as-a-judge evals catching >80% of codegen failures .
- Deployed scalable asynchronous API using job IDs and evaluated AWS Bedrock for cost optimization .
Big Data • Geospatial Analysis • Prescriptive Analytics • Clustering
- Processed 5.4M+ Aadhaar records across 1,045 districts, extracting operational and demographic features[cite: 10].
- Engineered district embeddings and unsupervised clustering to identify high-stress archetypes[cite: 10].
- Developed priority scoring engine and similarity-based district retrieval dashboard for policy prioritization[cite: 10].
RAG • Prompt Engineering • Guardrails • Adaptive Scheduling
- Built multi-turn viva system adapting difficulty in real-time .
- Added prompt-injection protection, schema validation, and spaced-repetition scheduling .
- Designed structured question generation using hint-first scaffolding .
Recommender Systems • Ranking • LLM Distillation • LoRA
- Finetuned Qwen2.5 models using LoRA for semantic tagging and cold-start handling .
- Implemented cart-prefix Top-N ranker yielding HitRate@10=0.718 and NDCG@10=0.398 .
- Surfaced 4,494 unique items with 0% duplicate recommendations at p99=52.08ms latency .
OpenCV • Object Detection • Flask • MJPEG Streaming
- Created real-time retail surveillance pipeline for live anomaly detection .
- Achieved stable near real-time performance with temporal validation and confidence thresholding .
- President, E-Cell — leading entrepreneurial and innovation initiatives.
- AI/ML Domain Lead, GDG Nagpur — leading workshops and mentoring projects in AI/ML.
- Data Analyst, CRISPR (IIIT Nagpur) — supporting hackathons, analytics demos, and student projects.
- UIDAI Aadhaar Hackathon — 2nd place among 20,000+ participants (Cash Prize ₹1,50,000).
- Coral Quest International Data Science Hackathon — global rank 6th (Cash Prize $600).