Skip to content
View ranji-t's full-sized avatar

Block or report ranji-t

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
ranji-t/README.md

Hi, I'm Ranjith Rajmohan 👋

AI/ML Engineer based in Pune, India. I build production-grade RAG systems, agentic pipelines, and ML systems — with a strong bias toward understanding the mathematics before touching the code.


🚀 Projects

CPU-only agentic RAG — no LangChain wrappers, no GPU, no cloud. LangGraph orchestration with adaptive web-search routing, conversational summarization, and query enrichment. Hybrid retrieval: dense (nomic-embed) + BM25, vectorized MMR via einsum, RRF fusion, full chunk provenance tracking. Gemma4 E2B via Ollama.

End-to-end fraud detection — JAX neural nets from scratch, multi-objective Optuna (Pareto recall vs precision), custom Precision@K / Recall@K / Lift@K metrics, SHAP explainability, probability calibration, and GitHub Actions CI/CD. No framework shortcuts.

Local-first RAG pipeline — ChromaDB, DeepSeek-R1 via Ollama, idempotent SHA-256 ingestion, reactive Marimo UI. Zero cloud APIs. Fully Dockerized. The foundation that led to the Gemma4 system above.


🔧 Stack & Engineering Discipline

LLM & Agents · LangGraph · LangChain · Ollama · RAG · Hybrid Retrieval
ML & Causal · JAX · XGBoost · scikit-learn · Optuna · Stan · Double ML · Shapley Values
Data · Polars · Pandas · SQL · Dagster · Databricks
Infra · FastAPI · Docker · Qdrant · ChromaDB
Languages · Python · SQL · R · learning Rust & Dart

All projects are linted with Ruff, dependencies managed via uv, tested with pytest, and quality-gated through GitHub Actions CI/CD — same discipline across personal projects as production work.


💼 Day Job

At Dentsu I build Media Mix Models — Bayesian and Frequentist, prior elicitation to production inference. Causal methods (S/T-learner, Double ML, Shapley values) to understand why media drives outcomes. JAX-accelerated MMM components for speed. Automated refresh pipelines in Databricks. Statistics is not a step in the process — it's how I read whether a model is telling the truth: AIC/BIC, Durbin-Watson, VIF, log-probability, t-values.


📌 What I'm Working On

  • RAGAS evaluation for the Gemma4 RAG pipeline
  • Observability stack: OpenTelemetry · Prometheus · Grafana
  • Exploring AI Engineer & Data Scientist roles in Pune

📫 Connect

LinkedIn GitHub

Pinned Loading

  1. Gemma4-langgraph-Local Gemma4-langgraph-Local Public

    CPU-only agentic RAG — no LangChain wrappers, no GPU, no cloud. LangGraph orchestration with adaptive web-search routing, conversational summarization, and query enrichment. Hybrid retrieval: dense…

    Jupyter Notebook 1

  2. fraud-detection-ml-system fraud-detection-ml-system Public

    End-to-end fraud detection — JAX neural nets from scratch, multi-objective Optuna (Pareto recall vs precision), custom Precision@K/Recall@K/Lift@K metrics, SHAP explainability, and GitHub Actions C…

    Jupyter Notebook 1

  3. naive-rag naive-rag Public

    Local-first RAG pipeline — ChromaDB, DeepSeek-R1 via Ollama, idempotent ingestion, reactive Marimo UI. Zero cloud APIs. Fully Dockerized.

    Python 1

  4. RAG-APP RAG-APP Public

    A full-stack RAG application combining a FastAPI/LangChain backend with a sleek Flutter frontend. Integrates Qdrant for vector search, Ollama for local embeddings, and GPT-4o for generation. Fully …

    Python 1