Skip to content
View kush0o7's full-sized avatar

Block or report kush0o7

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
kush0o7/README.md

Kushagra Tandon

CS @ Arizona State University · Graduating May 2026
I build AI agents, break LLMs for a living, and occasionally predict electricity prices

LinkedIn Portfolio Email


Featured Projects

Adversarial evaluation harness for LLM agents with automated safety grading

  • Proposer → Critic → Revision debate pipeline to expose model failure modes
  • Automated metrics: hallucination rate, refusal accuracy, policy compliance
  • Async evaluation via Redis/RQ with full trace storage in Postgres
  • Prompt-injection detection and controlled execution environments
  • Tech: Python · FastAPI · PostgreSQL · Redis · RQ · Docker

Full-stack AI safety application enforcing configurable rules on LLM outputs

  • Classifier-based pre/post generation gating with configurable harm thresholds
  • Model-independent rule engine preventing jailbreak overrides
  • Red-team evaluation pipeline with structured trace logging
  • Tech: Python · FastAPI · React · Scikit-learn · Docker

Production-grade AI system for voice-driven patient scheduling

  • Real-time chat and voice workflows with hallucination-safe backend validation
  • Deployed on AWS EC2 with Nginx, Redis session persistence
  • Tech: Node.js · TypeScript · React · Vapi.ai · Groq · Redis · AWS

Quantile forecasting model for EU electricity markets

  • LightGBM q10/q50/q90 model — 60% reduction in MAE vs baseline
  • Integrated Anthropic API with JSON audit logs and hard fallback paths
  • Tech: Python · LightGBM · Anthropic API · Pandas

Multi-tenant backend for organizational intent inference

  • Semantic drift detection using embeddings to surface early signals
  • Tenant-scoped REST APIs with pgvector similarity and timeline queries
  • Tech: FastAPI · PostgreSQL · pgvector · SQLAlchemy · Docker

Macro scenario engine with probabilistic modeling

  • Monte Carlo simulations across geopolitical and economic scenarios
  • Integrated FRED and GDELT data sources with 1.03% MAE validation
  • Tech: React · TypeScript · Python · Monte Carlo

Agent orchestration framework focused on measurable outcomes

  • Tech: Python

Calibrated sports outcome predictor

  • XGBoost classifier · ROC-AUC 0.79 on 570 out-of-sample matches
  • Feature engineering with odds-derived metrics and ELO ratings
  • Tech: Python · XGBoost · Scikit-learn · Streamlit

Systematic trading strategy with walk-forward validation

  • SMA crossover with ATR-based stops and portfolio simulation
  • Backtested across stocks, ETFs, and crypto
  • Tech: Python · Backtrader · yfinance

Skills

Languages: Python · C++ · TypeScript · Java · SQL
ML/AI: LightGBM · XGBoost · Scikit-learn · TensorFlow · Anthropic SDK · pgvector
LLM Systems: Adversarial evaluation · Red-team pipelines · Safety gating · Prompt-injection mitigation
Backend: FastAPI · Flask · Django · Redis · PostgreSQL · Docker · REST APIs
Frontend: React · TypeScript
Tools: Git · Docker · CI/CD · SQLAlchemy · Backtrader · AWS


Experience

Teleoperation Associate · Objectways Technologies LLC (2026 – Present)
Collecting and recording robot demonstration data using Trossen systems for ML training pipelines

Research Aide · ASU School of Earth & Space Exploration (Oct 2023 – May 2025)
Built data pipelines and virtual field modules for NSF-funded WORM Portal

Software Dev Intern · Hindustan Aeronautics Limited (Jun–Aug 2023)
Built internal dashboards for real-time aerospace telemetry monitoring


Open to roles in LLM evaluation, AI safety, ML engineering, quant systems, and forward deployed engineering

Pinned Loading

  1. Bet_Model Bet_Model Public

    End-to-end EPL model comparing XGBoost predictions with bookmaker odds, including calibration, backtests, and a Streamlit dashboard.

    Python 1

  2. Momentum_Following_Strategy Momentum_Following_Strategy Public

    A momentum strategy using the 50-day and 200-day SMA, buying on a Golden Cross and selling on a Death Cross. Backtested with Backtrader and yfinance, it works for stocks and crypto.

    Python 1

  3. Constitutional-Safety-Agent Constitutional-Safety-Agent Public

    A self-critiquing, rule-enforcing AI agent that produces outputs aligned with a defined internal constitution

    Python 1

  4. QuantAI QuantAI Public

    Built a multi‑tenant intent‑inference platform with FastAPI + Postgres (pgvector). It ingests job posts and filings, computes semantic drift, and generates evidence‑backed intent hypotheses. Added …

    Python 1

  5. Redline-Agent-Safety-Eval Redline-Agent-Safety-Eval Public

    Redline is a secure evaluation platform that stress‑tests LLM agents with adversarial cases and summarizes safety performance.

    Python 1

  6. WorldweLivein WorldweLivein Public

    Global macro forecasting engine — live economic signals, historical analogy matching, Monte Carlo scenario modeling, and geopolitical risk analysis.

    TypeScript 2