Ask a question about any of 799 SEC 10-K filings and get a cited answer. Hybrid search in Postgres (dense + BM25), fused and reranked, with each step kept only if it raised recall.
Read the case study →RAGHybrid SearchEvalsPython
Abhishek Tuteja
AI Engineer
Most of my day goes into breaking models and building evals that check the small details, like the right citation in PROVA or the right filing year in RAG-SEC. Off the clock I read (right now, Nexus), journal, meditate, and try new AI tools the week they ship, lately Jev and Dots.
I finished my MSCS at Northeastern in May 2026. Before that, I spent two years at Deloitte running daily risk analysis for a Fortune 100 commodities firm, checking 10,000+ transactions a day and tracing each break to its root cause. Evals feel like the same job, just for models.
Projects
Upload model risk documentation and get an SR 11-7 compliance score, the gaps, and the evidence for each one. Three agents, one per SR 11-7 pillar, check 20 elements in parallel.
Read the case study →Multi-AgentLLM-as-JudgeEvalsTypeScript
Experience
Jan 2025 - Apr 2026
Teaching Assistant
Northeastern University · Oakland, CA
- TA for three core CS courses: Intro to Programming (Python), Object-Oriented Design (Java), and Database Design (SQL).
- Ran graded code walks and office hours, and kept students' Docker-based database environments running for CS 3200.
May 2022 - Jul 2024
Advisory Analyst
Deloitte · Bengaluru, IN
- Ran the daily mark-to-market risk report for a Fortune 100 agricultural commodities client, reconciling thousands of positions and P&L records in SAP ACM and BW.
- Validated over 10,000 transactions a day. Traced breaks through the contract, load, and settlement chain to the root cause, so one fix cleared every similar error. Scoped each fix with developers and tested it before release.
- Led end-of-day reporting for a three-analyst team and supported the system's rollout across LATAM, EMEA, and APAC. Deloitte Spot Award.
Oct 2020 - Nov 2020
Machine Learning Intern
Steinn Labs · Remote
- Built an OCR pipeline in Python and TensorFlow that pulled structured fields from scanned ID documents.
- Raised extraction accuracy by about 15% through preprocessing and tuning, checked against labeled samples.