Balamurugan P G

A portfolio-allocation policy trained only on synthetic markets, then evaluated zero-shot on six real universes with 10-seed confidence intervals. PPO, no real prices during training. Research by Balamurugan P G and Jason Pandian. Ongoing project; does not claim to beat buy-and-hold (tested tie with SPY out of sample and behind SPY on the holdout point estimate).

Python package that detects train/eval data contamination with 4 checks: exact match, 13-gram overlap, fuzzy match and semantic similarity. Published on PyPI: pip install verascan.

OpenAI-compatible LLM gateway. A local logistic classifier scores the prompt, then conformal risk control chooses the cheap or strong model without embeddings or an extra model call for the decision. Offline frozen pairs on GSM8K, ARC-Challenge, and MBPP (3,188 pairs) achieved 32.77% estimated cost savings and 4.769% mean regret (length-shift replay regret falls from 20.215% to 4.799% after recalibration). Live Groq + xAI recorded 15.96% savings and 12.12% mean regret, missing the 5% target. Status: research prototype, GitHub only.

Multi-agent system for Indian IPO analysis. Every fact is cited to a page, then checked by a deterministic fact-checker; results are delivered on Telegram. 568 tests.

5-class chest X-ray classifier built on DenseNet121, with Grad-CAM heatmaps. 86–88% test accuracy. Full-stack app (FastAPI backend, React frontend, Supabase) deployed on GCP, live at medrag.in.

Pneumonia detection from chest X-rays with MobileNetV2. 92.5% test accuracy. Live Streamlit app.

Agent that solves LeetCode problems using a 5-provider LLM fallback chain and 4 Playwright browser workers. 230+ accepted.

Customer churn prediction dashboard built in Streamlit, trained with Gradient Boosting on 7,043 customers.

Cancer analytics dashboard in Streamlit over 17,686 records, with statistical analysis plus Gradient Boosting regression and classification (GBR/GBC).

Streamlit Cloud SLA monitor for U23CS702 with seeded simulation and optional read-only Google Cloud Monitoring. Tracks four SLA metrics: uptime, latency, error rate, and availability across light and black themes. Course project with a live Streamlit app.

Recurrent neural network (RNN) on household power consumption data.

Artificial neural network (ANN) that predicts Titanic passenger survival.

Hierarchical clustering on a height–weight dataset.

DBSCAN density-based clustering on a height–weight dataset.

K-Means clustering on a height–weight dataset.

Exploratory data analysis of video game sales.

Luxury car dealership web project.