Tarun Basavaraju
Software Engineer · Full-stack & Applied AI
Bengaluru+91 98458 59937tarunbasavaraju25@gmail.comLinkedInPortfolioGitHub
Summary
Final-year CSE student who ships production apps for a paying client and LLM systems measured by rigorous evaluation.
Experience
Freelance Software Engineer · EuroKids Preschool · ek-desk.vercel.app
Aug 2026 – Present · Bengaluru
Next.js · TypeScript · Supabase · Vercel
- Designed and launched EK Desk solo, replacing handwritten registers and Excel with a fee-management platform now running at 2 branches for 300+ students.
- Engineered expiring, tokenized payment links sent over WhatsApp that open directly in UPI apps, with a verification queue, approval workflow and audit log tracing every rupee.
- Automated dues calculations and balance tracking, saving staff 2-4 hours per billing cycle; added one-click overdue lists plus daycare re-enrollment, a front-desk request that trimmed data entry.
- Secured all personal and financial records with default-deny PostgreSQL Row Level Security, and verified transaction logic with 61 test suites (Vitest, database integration, Playwright E2E).
Product Associate Intern · Docxster
Jul 2024 · Bengaluru
Document AI startup
- Reduced dashboard load time from 5s to 2s (60% faster) by caching frequently requested values.
- Enhanced database schemas with new data types and revised the privacy policy for GDPR compliance.
- Prepared 1,000+ annotated bills and invoices to train the company's extraction model.
Projects
- ThirdEye: a smart-contract auditor that runs on your laptop
React · FastAPI · Ollama · ChromaDB
- Built and owned the verdict engine (router, evidence gate, risk pooling), cutting false alarms from 72.2% to 45.9% at 89.1% recall across 2,142 smart contracts.
- Created a local Solidity vulnerability scanner with 3 teammates, sending each file through 8 specialist LLM roles across 3 model families at $0 per scan.
- Analyzed how false-positive rates climb from 12.2% to 90.1% as routing broadens, replicating the trend on a held-out seed; manuscript in preparation.
- ADA: an adaptive RAG assistant for research papers
Python · FAISS · Ollama · Streamlit
- Developed an adaptive RAG assistant for research literature with 2 classmates; it classifies queries, retrieves over MiniLM embeddings and self-critiques its answers.
- Implemented an LLM-as-judge scoring harness on a 100-question benchmark; against a naive baseline, ADA lifted faithfulness from 63% to 84% and its correctness score by 35%. Co-authored an IEEE-format paper.
Skills
- Languages
- TypeScript · JavaScript · Python · Java · C++ · C · SQL
- Frontend
- React · Next.js · Tailwind CSS · Streamlit
- Backend
- Node.js · FastAPI · Supabase · REST APIs · Zod
- Databases
- PostgreSQL (RLS, migrations) · FAISS · ChromaDB
- AI/ML
- RAG · LLM-as-judge evaluation · prompt engineering · embeddings · Ollama
- Testing & DevOps
- Vitest · Playwright · Docker · Git · Linux · Vercel
- Core CS
- DSA · OOP · DBMS · Operating Systems · Computer Networks
- Familiar
- Spring Boot · AWS · Kafka · Spark
Education
B.Tech, Computer Science & Engineering · PES University, EC Campus
2023 — 2027
Bengaluru
Class 12 (PCMC) · Delhi Public School, Bangalore North
Bengaluru
Leadership & Achievements
Trustee, The Celestial Foundation (registered NGO)
Feb 2023 – Present
Directed planning, procurement and logistics for 6+ community projects reaching 1,800+ people, including a 3-day Daan Utsav that donated 100+ kg of rations to 2 orphanages.
Top 10, Override’23 Hackathon
2023 · GDSC, PES University