👋 Hi, I’m Aditi Khare — AI researcher, builder, and open-source contributor.
AI Research Junction is a curated, continuously updated collection of high-impact AI research papers, paired with production-grade Generative AI system design and cloud architecture insights (AWS-first reference).
This repository is built for readers who want to:
- stay current with AI research, and
- understand how ideas translate into real, scalable systems
🚀 Read less. Understand more. Build better AI.
⭐ If this repository helps you stay current with AI research and system design, consider starring it.
If you care about:
- 🔬 Cutting-edge AI research (LLMs, GenAI, Agents, CV, Quantum AI)
- 🏗️ Production-grade Generative AI systems
- ⚙️ Inference efficiency, cost, reliability & safety
- ☁️ AWS-based cloud-native AI architectures
- 📚 Signal-over-noise curation you’ll return to
This repository complements my ongoing AI research newsletter and curated insights.
Aditi Khare
- AI Researcher & Open-Source Contributor
- Enterprise AI Leader — Product, Infrastructure & AI Architecture
- Principal ML Scientist | AI Product Owner & Architect
- AWS & AI Research Expert | Author
Research Interests
🔷 Generative AI · 🔷 Agentic AI · 🔷 Computer Vision · 🔷 Quantum AI
Most repositories list papers.
AI Research Junction focuses on judgment, context, and systems thinking.
Each paper or resource is curated with attention to:
- 🧠 Core research contribution
- 🔁 Practical relevance
- ⚙️ Inference & deployment implications
- 🧪 Evaluation, robustness & failure modes
- 🚧 Where research breaks in real systems
- Prompt design for reliability & consistency
- Retrieval-Augmented Generation (RAG) architectures
- Agentic workflows & tool orchestration
- Memory, context windows & long-horizon reasoning
- Evaluation, guardrails & safety layers
- Cost-aware inference & model routing
This repository uses AWS as a reference architecture for scalable, secure, production-grade AI systems.
- Separation of inference, retrieval & orchestration layers
- Stateless model serving with scalable backends
- IAM-first security & least-privilege access
- Observability by default (metrics, logs, traces)
- Cost controls, quotas & graceful degradation
📘 AWS Well-Architected Framework (AI/ML)
- https://docs.aws.amazon.com/wellarchitected/latest/framework/welcome.html
- https://docs.aws.amazon.com/wellarchitected/latest/machine-learning-lens/welcome.html
- Amazon EKS – Kubernetes for model serving & agents
https://docs.aws.amazon.com/eks/latest/userguide/what-is-eks.html - Amazon ECS / Fargate – Managed container inference
https://docs.aws.amazon.com/AmazonECS/latest/developerguide/Welcome.html - AWS Lambda – Lightweight orchestration & glue logic
https://docs.aws.amazon.com/lambda/latest/dg/welcome.html
- Amazon SageMaker – Training, inference & pipelines
https://docs.aws.amazon.com/sagemaker/latest/dg/whatis.html - Amazon Bedrock – Managed foundation models (GenAI)
https://docs.aws.amazon.com/bedrock/latest/userguide/what-is-bedrock.html - AWS Step Functions – Agent & workflow orchestration
https://docs.aws.amazon.com/step-functions/latest/dg/welcome.html
- Amazon S3 – Feature, document & dataset storage
https://docs.aws.amazon.com/AmazonS3/latest/userguide/Welcome.html - Amazon OpenSearch Service – Vector & hybrid search
https://docs.aws.amazon.com/opensearch-service/latest/developerguide/what-is.html - Amazon DynamoDB – Session state & agent memory
https://docs.aws.amazon.com/amazondynamodb/latest/developerguide/Introduction.html
- Amazon CloudWatch – Metrics, logs & alarms
https://docs.aws.amazon.com/cloudwatch/index.html - AWS IAM – Identity & access management
https://docs.aws.amazon.com/IAM/latest/UserGuide/introduction.html - AWS KMS – Encryption & key management
https://docs.aws.amazon.com/kms/latest/developerguide/overview.html
- RAG pipelines with decoupled retrieval & inference
- Agent-based systems using Step Functions + tools
- Async inference for latency-sensitive paths
- Batch + real-time hybrid pipelines
- Multi-model routing & fallback strategies
Strong GenAI systems are designed, not improvised.
| Paper | Focus | Link | Category |
|---|---|---|---|
| Copyright Detection in LLMs | Memorization risks | https://arxiv.org/abs/2511.20623 | Generative AI |
| Beyond Automation – Governance in GenAI | Governance & work | https://arxiv.org/abs/2512.11893 | Generative AI |
| AgentEval | Evaluating agents | https://arxiv.org/abs/2512.08273 | Agentic AI |
| Paper | Focus | Link | Category |
|---|---|---|---|
| GenAI × Extended Reality | XR + GenAI | https://arxiv.org/abs/2511.03282 | Generative AI |
| GenAI in Qualitative Research | Methods & risks | https://arxiv.org/abs/2511.08461 | Generative AI |
| Safety Guardrails | Alignment & safety | https://arxiv.org/abs/2511.15732 | Generative AI |
| Paper | Focus | Link | Category |
|---|---|---|---|
| Generative AI – Deep Survey | Models & use cases | https://arxiv.org/pdf/2510.21887 | Generative AI |
| GenAI & Scientific Writing | Empirical study | https://arxiv.org/abs/2510.17882 | Generative AI |
| Chronologically Consistent GenAI | Temporal consistency | https://arxiv.org/abs/2510.11677 | Generative AI |
| Paper / Model | Focus | Link | Category |
|---|---|---|---|
| DeepSeek-V3 | Open LLM | https://github.com/deepseek-ai/DeepSeek-V3 | Generative AI |
| Inference-Time Self-Improvement | Self-refining LLMs | https://arxiv.org/pdf/2412.14352 | Generative AI |
| Modern BERT | NLP advances | https://arxiv.org/abs/2412.13663 | NLP |
| Paper / Tool | Focus | Link | Category |
|---|---|---|---|
| OpenAI Swarm | Multi-agent workflows | https://github.com/openai/swarm | Agentic AI |
| Claude 3.5 | Reasoning & multimodality | https://www.anthropic.com/news/3-5-models-and-computer-use | Generative AI |
| Paper / Tool | Focus | Link | Category |
|---|---|---|---|
| Llama 3.2 | Edge AI & vision | https://www.llama.com | Edge AI |
| Self-Correction via RL | Reasoning | https://arxiv.org/abs/2409.12917 | Generative AI |
| Iteration of Thought | Inner dialogue | https://arxiv.org/abs/2409.12618 | Generative AI |
| OpenAI o1 | Reasoning models | https://openai.com/index/introducing-openai-o1-preview | Generative AI |
| AutoGen Studio | Agent orchestration | https://github.com/microsoft/autogen | Agentic AI |
| Strategic CoT | Advanced reasoning | https://arxiv.org/abs/2409.03271 | Generative AI |
| RAG Noise | Retrieval robustness | https://arxiv.org/abs/2408.13533 | Generative AI |
| GameGAN | Simulated worlds | https://github.com/nv-tlabs/GameGAN_code | Generative AI |
| Agentic RAG | Time-series RAG | https://arxiv.org/abs/2408.14484 | Generative AI |
| Paper | Focus | Link | Category |
|---|---|---|---|
| The AI Scientist | Automated discovery | https://paperswithcode.com/paper/the-ai-scientist-towards-fully-automated-open | Generative AI |
| ControlNeXt | Video & image control | https://arxiv.org/pdf/2408.06070v2 | Computer Vision |
| RAG-Checker | RAG diagnostics | https://arxiv.org/abs/2408.08067 | Generative AI |
| Paper | Focus | Link | Category |
|---|---|---|---|
| Microsoft SAMBA | Efficient LLMs | https://arxiv.org/pdf/2406.07522 | Generative AI |
| Quantinuum Quixer | Quantum transformers | https://openai.com/index/extracting-concepts-from-gpt-4 | Quantum AI |
| No Language Left Behind | Multilingual translation | https://github.com/facebookresearch/fairseq/tree/nllb | Generative AI |
| Paper / Model | Focus | Link | Category |
|---|---|---|---|
| Meta Llama-3 | Long-context LLMs | https://huggingface.co/papers/2404.19553 | Generative AI |
| Multi-Token Prediction | Faster inference | https://arxiv.org/abs/2404.19737 | Generative AI |
| Phi-3 | Small efficient LLMs | https://arxiv.org/pdf/2404.14219 | Generative AI |
| FrugalGPT | Cost-aware LLM usage | https://portkey.ai/blog/implementing-frugalgpt-smarter-llm-usage-for-lower-costs | Generative AI |
| GNN-RAG | Graph-based RAG | https://github.com/cmavro/GNN-RAG | Generative AI |
- Track AI research trends
- Discover models worth experimenting with
- Understand trade-offs before building
- Reference system & architecture decisions
⭐ If this repository is useful, consider starring it to keep it on your radar and support my ongoing research and curation.