Mukund Ojha

Forward Deployed Engineer, AI Engineer, and Machine Learning Engineer in the San Francisco Bay Area. I build AI agents that survive production. I build them for real customers, then deploy them myself.

I like the part of AI work that happens after the demo: getting a system into production, making it reliable, and making sure people keep using it. I started at Avaada Energy as the first dedicated ML hire on a new AI initiative, where solar forecasting saved over $50K a year and anomaly detection raised uptime 18% on a 500+ unit robot fleet. At UC Santa Cruz I moved into LLM systems, built GradeRight to 30,000+ users, and lead engineering at TopTutors. Based in the San Francisco Bay Area, looking for forward deployed, AI, software, and machine learning engineering roles from January 2027.

Email: muojha@ucsc.edu | LinkedIn | GitHub | GradeRight.org | Download résumé (PDF)

Experience

Founding Engineer, TopTutors (May 2026 to Present)

Lead a team of 4 engineers on a B2B school platform built 0 to 1 and used across 200+ schools, with 150+ zero downtime production releases.
Built an AI hiring pipeline that scores tutor interview transcripts against a versioned rubric, with admins able to correct any score with a reason.
Designed the data model, PostgreSQL with role based access across 7 user roles and per school data isolation enforced at every endpoint that writes data.
Built an LLM pipeline that ingests chunked PDFs, extracts questions, and autogrades submissions, cutting assessment digitization from hours to minutes.
Work directly with nontechnical school stakeholders to turn operational problems into features, then defend the technical tradeoffs back to them.
Own the full stack and the AWS infrastructure under it, Next.js and React on the front, Python and TypeScript services on PostgreSQL behind it.

Founder and Software Engineer, GradeRight.org (September 2025 to Present)

Built and run a production AI grading platform serving 30,000+ users as founder and sole engineer.
Plugs into Canvas over LTI 1.3 and grades 500 submissions in under 2 hours with the consistency of a trained marking panel.
Designed the agent pipeline behind it, a relevance gate, parallel grading passes against a structured rubric, and median aggregation that removes run to run drift.
Built the evaluation harness that gates every release, golden sets and automated accuracy checks, with instructor override held under 10% as ground truth.
Cut serving cost by routing light gating work to a small model and self hosting the heavy path on vLLM.
Hardened the pipeline against prompt injection on all student input and enforced deterministic decoding.

Assistant Manager, Machine Learning Engineer, Avaada Energy (September 2023 to September 2025)

Built solar generation forecasting for 32+ sites across India, cutting deviation penalties on the Indian Energy Exchange and saving $50K+ a year.
Built anomaly detection and preventive maintenance for a 500+ unit fleet of autonomous solar panel cleaning robots, raising uptime 18%.
Engineered Spark ETL pipelines over fleet telemetry and built the anomaly detection and forecasting models on top of them.
Deployed containerized services on Docker and Kubernetes across AWS and Azure, with monitoring, alerting, and on call recovery.

Work

GradeRight (founder and sole engineer, live since 2025). AI grading inside Canvas, consistent enough that instructors overrule fewer than one score in ten. 30,000+ users, 500 submissions graded in under 2 hours. Relevance gate, three parallel grading passes on self-hosted vLLM, median aggregation, instructor review, and a release gate of golden sets.
Commit Before Trust (multiagent orchestration, paper under review, 2026). Agents share one typed state, and an LLM judge checks every update before it commits, with semantic checkpointing to roll back bad writes. 62% fewer tokens than full history passing, HotpotQA 50% to 54%.
GradeINK. Grading for handwritten math that localizes each error on the page with a grounding vision language model, line segmentation, and IoU region snapping, with partial credit for errors carried forward.
CaseLumen. Agentic legal research over a 5,000+ document corpus, designed after surveying Indian lawyers, with every answer cited. Hybrid BM25 and dense search, reranking, LangGraph, and RAGAS evaluation. Recall up 22%, research time down 40%.
SlugMarket. A UC Santa Cruz student marketplace, live on the App Store. Verified campus sign in, Row Level Security, realtime chat, and push notifications on Supabase.
VagBAY (client work, in development). A marketplace and community app for Volkswagen, Audi and Skoda owners in India: React Native, NestJS, PostgreSQL, and a Next.js moderation dashboard, integrated with the client's existing accounts.
The Select Social (in development). An invite-only dating app for India's metro cities, one verified match at a time. SwiftUI over FastAPI, with hand-reviewed applications and selfie verification.

Research

Commit Before Trust: Transactional Shared Memory for Multi-Agent LLMs (Under Review, 2026). Mukund Ojha, Sakshi Konnur, Saranya Kolachana, UC Santa Cruz. 62% fewer tokens, HotpotQA 50% to 54%.
Diagnosing How Frontier LLMs Fail at Lean: A miniF2F-v2 Evaluation Study (Under Review, 2026). Mukund Ojha, sole author, UC Santa Cruz.
RAG Robots: Retrieval-Augmented Generation for Enhanced Robotic Task Planning (Under review, 2026).
Two Stage Residual Models for Robust Cloud Ramp Prediction (Under review, 2025).

Expertise

AI agents and orchestration (LangGraph, CrewAI, AutoGen, tool calling, checkpoint and rollback). Agent evaluation and reliability (golden set regression harnesses, LLM as judge, RAGAS, human in the loop review). LLM systems in production (LLM APIs, prompt engineering, vLLM serving, model routing, prompt injection hardening). Retrieval and RAG (FAISS, Qdrant, ChromaDB, hybrid BM25 and dense search, cross encoder reranking). Forward deployed delivery (scoping with nontechnical stakeholders, pilot through production rollout, incident response). Backend, data, and infrastructure (Python, TypeScript, FastAPI, React, PostgreSQL, Spark, PyTorch, Docker, Kubernetes, AWS, Azure, CI/CD).

Education

M.S. in Computer Science and Engineering, University of California Santa Cruz, GPA 3.92/4.00, graduating December 2026. B.Tech in Computer Science Engineering, Chitkara University, GPA 4.0/4.0.