Skip to content
View aditya-ag26's full-sized avatar

Highlights

  • Pro

Block or report aditya-ag26

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
aditya-ag26/README.md

Aditya Agarwal

AI/ML Engineer  ·  LLM Serving  ·  RAG  ·  Graph Learning

LinkedIn Email Resume

🔬  AI Research Intern @ SBILab, IIIT Delhi — LLMs for clinical docs, GNNs for drug repurposing
🎓  B.Tech CSE, minor in AI/ML — GGSIPU Delhi
⚡  I like the part after the notebook: serving, latency, cost
📬  Open to AI/ML engineering roles

Things I've built

Highlight
Meridian — KOL influence platform
Resolves researcher identity across 4 scholarly APIs, scores 14 influence indicators. Human-in-the-loop LangGraph runs that pause and resume mid-pipeline.
FastAPI LangGraph pgvector AWS
Same-field similarity 0.65
vs cross-field 0.14
AI Consciousness Assistant — cited RAG over 158 papers
A calibrated guardrail (F1 0.939) declines out-of-scope questions before any model call. Two autoscaling Cloud Run services behind CI/CD.
FastAPI FAISS ONNX GCP
Cold start 110s → 19s
via PyTorch → ONNX
Smart Stock Predictor — LSTM + Transformer forecasts
Deploys gate on tests, Trivy scans, and CodeQL. Squeezed onto a 512 MB tier with an LRU model cache and worker recycling.
Django TensorFlow React Docker
MAPE 3.41%
Weak Signals in CV — research trend forecasting
Temporal keyword pipeline building Emergence and Impact Maps to surface research frontiers.
Python Pandas NLP
13 years of arXiv data

At SBILab

Clinical documentation — Entity-aware generation via SciSpacy NER set differences. LLaMA-3-8B, Mistral-7B and Qwen3-4B served through vLLM batched inference, hierarchical RAG on Qdrant.  →  Rouge-1 F1 0.543

Drug repurposing — GATv2 over gene PPI networks with a BPR reversal prior, LangGraph agents fusing multi-source evidence.  →  Recall@10 +23% across 5 cancers

Stack

Languages Python SQL JavaScript C++
ML PyTorch PyG TensorFlow ONNX OpenCV
LLM & Retrieval vLLM HuggingFace LangGraph Qdrant pgvector
Backend FastAPI Django PostgreSQL pytest
Cloud & Ops AWS GCP Docker Actions

What I care about

Models are only half the job.

Build Measure Ship
RAG pipelines that actually retrieve the right context Retrieval quality, model accuracy & hallucination Containerized services with CI/CD
LLM inference & serving Latency, throughput & cold starts Cloud deployments that survive real traffic
Graph-based ML systems Ranking, recall & calibration Systems that are observable and maintainable

Pinned Loading

  1. ai-consciousness-project ai-consciousness-project Public

    An end-to-end research pipeline and chatbot system that explores AI Consciousness through scholarly papers, keyword analysis, and RAG-based LLMs. Includes arXiv scraping, metadata filtering, FAISS …

    Python

  2. fisheye-correction fisheye-correction Public

    Fisheye distortion correction

    Python

  3. KOL-influence-scoring-system KOL-influence-scoring-system Public

    An AI-powered platform that scores domain experts using information extracted from heterogeneous public sources for similarity and influence analysis.

    Python

  4. pnp pnp Public

  5. Smart-Stock-Prediction Smart-Stock-Prediction Public

    Python

  6. sidsharmaa/weak-signals-new sidsharmaa/weak-signals-new Public

    Jupyter Notebook 1 1