AI/ML Engineer with hands-on experience building LLM systems, RAG pipelines, and backend services across industry and academic internships. Strong foundation in ML, DL, and GenAI, grounded in first-principles understanding, with demonstrated ability to evaluate LLM behavior and ship reliable infrastructure. Currently interested in designing reliability-aware, agent-ready GenAI systems with robust evaluation, monitoring, and scalable infrastructure.
Master's in Data Science — Defence Institute of Advanced Technology (DIAT), Pune
Bachelor's in Mathematics — Indira Gandhi National Open University (IGNOU), New Delhi
| Aug 2025 Jan 2026 |
Founding Engineering Team · Observo Tech Studio Benchmarked email service providers, warm-up tools, and video analytics APIs for an AI-powered B2B marketing product. Built the early backend for multi-mailbox outbound campaigns with conversation-level state across 5 mailboxes (FastAPI + PostgreSQL). Implemented a real-time webhook pipeline capturing 16+ delivery and engagement signals. Deployed serverless APIs on Azure Functions and Logic Apps. |
| Apr 2025 Jul 2025 |
GenAI Engineer, R&D · Confedo AI Evaluated 4 LLM observability frameworks by implementing equivalent RAG workflows and comparing tracing quality, async context propagation, and metric reliability. Reverse-engineered tracing implementations across 3 open-source codebases. Prepared experimental RAG pipelines assessing retrieval quality and faithfulness using LLM-based and classical IR metrics. |
| Jan 2025 Jun 2025 |
Research Intern · Usable Security Group (USG) Lab @ IIIT Delhi Designed a zero-shot machine unlearning framework enabling class-level forgetting without original data. Used contrastive model inversion to generate 5K synthetic samples for targeted unlearning. Achieved 90%+ class-wise unlearning efficacy on SVHN with 77% runtime reduction vs. GKT. Benchmarked against 4 baselines. Explored coreset based strategies for effective unlearning. |
| Languages |
|
| ML / DL |
|
| LLM / RAG |
|
| Backend |
|
| Cloud |
|
| Data |
|
|
Structure-aware documentation RAG pipeline built from scratch. Crawls 104 FastAPI pages, preserves section hierarchy, generates 971 metadata-rich chunks, and delivers citation-grounded answers via Pinecone + Groq.
|
RAG system with SHA-256 document deduplication, namespace isolation, semantic reranking, and inline-cited answers. Upload a 200-page manual, ask a question, get a precise cited response.
|
|
Fully local, zero-cloud RAG system. Chat with your own PDFs using Ollama + Streamlit. Ships with a privacy-law corpus (GDPR, CCPA, LGPD, DPDP) for interactive legal research.
|
GPT-style decoder-only Transformer implemented from first principles in pure PyTorch. 18.9M-parameter model with ablation studies on Pre-LN vs Post-LN, sinusoidal vs RoPE, tracked via W&B.
|
|
Modular computer vision pipeline using YOLOv8 for object detection and pose estimation. Classifies postures (standing, sitting, bending) via geometric heuristics on skeletal keypoints.
|
Sequential frame classification for gesture recognition using CNN feature extraction + LSTM temporal modeling. Built for TV gesture control with 5 distinct hand gesture classes.
|
> Agentic systems design and evaluation workflows
> LLM observability and tracing infrastructure
> Production RAG with evaluation (RAGAS), deduplication, and structured retrieval
> Machine unlearning — zero-shot frameworks for class-level forgetting
