ONLINE LOC: NEW DELHI ROLE: AI_ENGINEER VIEWS: traffic

AMAN
MISHRA

Hands-on with LLM systems, RAG pipelines, and production AI infrastructure. Strong foundation in ML, DL, and GenAI — engineering first, always.

LINKEDIN GITHUB RESUME CONTACT
EXPERIENCE
AUG 2025 — JAN 2026 // REMOTE
Observo Tech Studio
Founding Engineering Team

Built an MVP email infrastructure supporting sending, receiving, and replying across multiple mailboxes, designed as an internal tool for a larger agent-based system. Implemented real-time event ingestion pipelines to capture 16+ delivery and engagement signals. Developed modular analytics for granular engagement insights. Refactored backend APIs using FastAPI + JWT auth, deployed on Azure Functions with Logic Apps orchestration.

APR 2025 — JUL 2025 // REMOTE
Confedo AI
GenAI Engineer, R&D

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. Built prototypes using LangChain to test RAG evaluation metrics and execution flows.

JAN 2025 — JUN 2025 // NEW DELHI
USG Lab @ IIIT Delhi
Research Intern (Master's Thesis)

Developed a zero-shot machine unlearning method using contrastive model inversion, enabling privacy-preserving targeted forgetting without access to training data. Achieved 90%+ unlearning efficacy across 8 SVHN classes with 77% runtime reduction vs. GKT baseline. Benchmarked against 4 established baselines (EMMN, zMuGAN, GKT, Retraining).

STACK

AI / ML / DL

PyTorch Transformers PEFT scikit-learn

LLM Systems

LangChain LangGraph LangSmith RAG

Backend & Data

FastAPI PostgreSQL MongoDB Supabase Pinecone FAISS ChromaDB

Cloud & Infra

Azure Functions Azure Logic Apps Blob Storage Docker Git

Tracking

Weights & Biases
EDUCATION
2023 — 2025

MSc Data Science

DIAT, Pune
2018 — 2021

BSc Mathematics

IGNOU
PROJECTS VIEW ALL →

DocStructRAG

SOURCE

Structure-aware documentation RAG pipeline with custom ingestion, hierarchical chunking, and citation-grounded answers.

RAG NLP LLM

Precision RAG

SOURCE

High-precision RAG with SHA-256 deduplication, namespace isolation, semantic reranking, and inline-cited answers.

RAG DEDUPLICATION VECTOR SEARCH

Offline RAG Bot

SOURCE

Fully offline RAG system using local LLMs via Ollama for privacy-preserving document Q&A with Streamlit UI.

LOCAL LLM PRIVACY RAG

YOLO Posture

SOURCE

Modular CV pipeline leveraging YOLO detection and pose estimation with geometric posture classification.

COMPUTER VISION YOLO OPENCV

GPT Style Transformer

SOURCE

Decoder-only transformer from first principles in pure PyTorch. 18.9M params with Pre-LN/RoPE ablation studies.

PYTORCH TRANSFORMERS DEEP LEARNING

CNN-LSTM

SOURCE

Hybrid CNN+LSTM architecture for sequential frame classification. Built for gesture recognition on video data.

COMPUTER VISION TIME SERIES CNN-LSTM

Kolmogorov-Arnold Networks

SOURCE

Comparative study of classical ML, neural networks, and Kolmogorov-Arnold Networks on classification tasks.

KAN NEURAL NETS RESEARCH

Deep ML Solutions

SOURCE

Implementation of foundational ML algorithms, linear algebra routines, and deep learning architectures from scratch.

DEEP LEARNING MATH ALGORITHMS

Media Compression

SOURCE

Image and video compression using K-Means and Fuzzy C-Means clustering techniques.

COMPRESSION CLUSTERING MEDIA