Divyansh
Bansal

Engineer · Builder · Researcher. AI, agentic systems and whatever seems interesting enough to break.

The process

Ask the right question, stay with it, iterate until it holds, then make the useful thing real.

What I build

Questions become systems. Systems become things people can use.

• IntelliCode · Hybrid RAG

  Intelligence. Hybrid retrieval and AST analysis over a codebase. 121 tests · CI-gated. Finding the right code means matching exact identifiers and intent at the same time, and without measurement there is no way to tell which retrieval stage is actually helping. A hybrid RAG pipeline: dense FAISS and sparse BM25, fused with reciprocal rank fusion and reranked by a cross-encoder, answering with Qwen2.5-3B on Hugging Face ZeroGPU. Beside it, an async-aware AST analyser covering 12 anti-pattern classes, and a security scanner for injection, weak crypto and hardcoded secrets. A labelled evaluation harness (MRR@5, Recall@k, NDCG) measures each stage's lift, and 121 tests gate every change in GitHub Actions CI. Built with Python, FAISS, BM25, Cross-Encoder Reranking, Hugging Face Transformers, ChromaDB, CI/CD Eval Gates, GitHub Actions.

  IntelliCode · Hybrid RAG: CodeIntelliCode · Hybrid RAG: Live demo

• GPT From Scratch

  Models. A decoder-only transformer with no framework abstractions. 10.7M parameters · PyTorch. Using transformers every day is not the same as understanding one. The only way to be sure was to build it with nothing hidden. A decoder-only transformer of about 10.7M parameters in raw PyTorch: multi-head self-attention, positional encoding, layer normalisation and autoregressive generation, all written by hand. Training convergence and generation quality validated end to end on text corpora. Built with PyTorch, Python, Tokenisation, Sequence Modelling, Text Generation.

  GPT From Scratch: Code

• Loan Risk Intelligence

  Engineering. A stacking ensemble over ~1.8M LendingClub loans, SHAP-audited and drift-monitored. 0.9184 test AUC. Credit models trained on years of loans can quietly learn from the future. Across ~1.8M LendingClub loans, a 2016 underwriting regime shift leaks through any split that ignores time. A two-stage stacking ensemble (XGBoost and PyTorch) on year-stratified splits, audited with SHAP, and served from GCP Cloud Run (FastAPI, Docker) with live TreeSHAP explanations, a Streamlit dashboard and PSI/KS drift monitoring. A test AUC of 0.9184, validated with McNemar's test (χ² = 194.8, p < 0.0001). The API and the dashboard are live. Built with XGBoost, PyTorch, SHAP Explainability, MLflow, DagsHub, Drift Monitoring (PSI/KS), GCP Cloud Run, Docker, FastAPI, Streamlit.

  Loan Risk Intelligence: CodeLoan Risk Intelligence: APILoan Risk Intelligence: Dashboard

• DTU ERP Platform

  Product. An ML-powered ERP for higher education, live across the university. 1,200+ users. Preparing the university for accreditation took 25 days of manual work. An ML-powered ERP with 8+ modules for accreditation automation, analytics and networking, and a semantic-retrieval pipeline (vector embeddings, ChromaDB). I co-founded it and led the architecture, the frontend and adoption across stakeholders. Preparation went from 25 days to 7 (about 72% less). 1,200+ users across the university, a paper in Wiley's Software: Practice and Experience, and DTU funding for continued development. Built with ChromaDB, Vector Embeddings, Semantic Search, Docker, PostgreSQL.

  DTU ERP Platform: Paper · Wiley

• Five-Agent Reconciliation Platform

  Featured · Amdocs · AT&T. Five AI agents that took over reconciliation work on the AT&T account. Live in production. Reconciliation on the AT&T account ran on manual work every week, across systems that had to agree exactly. Five agents on a custom Python/FastAPI runtime with Azure OpenAI (GPT-4.1), each with its own cost and latency budget, deterministic fallbacks and human-in-the-loop checkpoints. Underneath: a unified async LLM gateway (structured output, bounded self-repair, circuit-breaker failover), retrieval-grounded NL-to-SQL with schema pruning and dry-run self-correction, a hermetic CI-blocking evaluation gate, and a four-plane agent memory (Redis, PostgreSQL, embeddings, a procedural library) that drives a self-improving correction loop. All five agents are live in production on the AT&T account, taking over manual reconciliation work. The architecture earned production sign-off from cross-functional stakeholders. Built with Python, FastAPI, Azure OpenAI (GPT-4.1), Agent Orchestration, NL-to-SQL, Structured Output Generation, Redis, PostgreSQL, Vector Embeddings, CI/CD Eval Gates.

Stack

• AI & LLM Engineering: Azure OpenAI (GPT-4.1), LangChain, LangGraph, MCP, RAG, Agent Orchestration
• Machine Learning & Deep Learning: PyTorch, TensorFlow, scikit-learn, XGBoost, LoRA / PEFT Fine-tuning, SHAP Explainability, Statistical Hypothesis Testing
• Retrieval & Search: FAISS, ChromaDB, Semantic Search, Vector Embeddings, BM25, Cross-Encoder Reranking
• Data & Infrastructure: Python, FastAPI, Docker, GCP Cloud Run, PostgreSQL, Redis, Apache Airflow
• Natural Language Processing: Hugging Face Transformers, NL-to-SQL, Text Generation, Structured Output Generation, Sequence Modelling, Tokenisation
• MLOps & Evaluation: MLflow, DagsHub, CI/CD Eval Gates, Drift Monitoring (PSI/KS), Model Versioning
• Tools & Languages: Git, GitHub Actions, Claude Code, GitHub Copilot, Pandas, NumPy, Streamlit

It started somewhere.

• 2021, The beginning: Started B.Tech at Delhi Technological University.
• 2022, First builds: Started turning ideas into systems.
• 2024, The ERP: 25 days → 7. We thought the process could be better. So we changed it.
• 2025, DTU ERP: 1,200+ users. A research paper. One university problem that got considerably larger.
• 2025, Amdocs · AT&T: AI software engineer on the AT&T account. Five agents, live in production.

Still building.

Agentic AI for AT&T, LLM systems, research, and whatever problem seems worth solving next.