Agents · Evals & Safety · RAG · Recommender Systems · Data Engineering
I'm an AI and Python engineer building practical intelligent systems: research agents, retrieval pipelines, graph ML, structured LLM workflows, and the data infrastructure behind them. I care about outputs that are explainable, testable, and useful beyond a demo.
A flight recorder for tool-using agents: adversarial suites, safety scoring, a human review console, and first-divergence trace diffing.Python · prompt-injection defense · SQLite · OpenTelemetry |
Short-video recommender: two-tower retrieval with FAISS, a multi-task ranker, and a LinUCB bandit. 1.55× the recall of a popularity baseline.PyTorch · FAISS · FastAPI · bandits · case study |
Claude streams short, speakable answers for live technical interviews, in four modes, each held to a spec by an LLM-judge eval.Claude API · FastAPI · SSE · evals · case study |
Web-search agent that must save typed findings and timeline events before it can submit a validated intelligence brief.React · Express · OpenAI Responses API · Zod |
Next.js + FastAPI + Chrome extension: job-fit scoring and bullet-level resume changes you approve, each backed by evidence.Next.js · FastAPI · Manifest V3 · case study |
Plaintiff-side legal AI: cited medical chronology, issue spotting, evidence coverage, a damages model, and grounded drafts.Node.js · document intelligence · citation grounding |
Every number traces to its source box and calculation, and a work queue is ranked by what matters now. Responsive, no dependencies.live demo · code |
Case studies, architecture notes and a recruiter AI assistant.
|
| Project | What it demonstrates |
|---|---|
| GenAI Toolkit | A tested Python package and CLI for chat, streaming, Responses API calls, and tool loops across OpenAI-compatible providers. |
| Drug Interaction Discovery | A GNN, biomedical knowledge graph, and literature-retrieval pipeline for explainable drug-interaction research. |
| Real-Time Voting Pipeline | Kafka, Spark Streaming, PostgreSQL, Docker, and Streamlit working together in an end-to-end data engineering system. |
| Real-Time AI Sentiment Streaming | Streams Yelp reviews through Spark, labels sentiment with an LLM, and publishes the results to Kafka. |
models PyTorch · GNNs · two-tower retrieval · bandits · structured generation
agents Claude & OpenAI APIs · tool calling · RAG · evals · guardrails
python data pipelines · CLIs · automation · Streamlit
data PostgreSQL · Kafka · Spark · knowledge graphs
delivery Docker · AWS · APIs · CI/CD
- Evaluating and red-teaming tool-using agents before they ship
- Multi-step AI agents with structured, inspectable outputs
- Retrieval and knowledge graphs for evidence-grounded reasoning
- Python systems that connect models, data, and real product workflows
Have an AI or Python problem worth solving?
akhilme008@gmail.com