Himadri Mishra

Senior AI Engineer

IIT-BHU  |  8 years building AI that ships

I work across the full AI stack, from computer vision and NLP to agentic LLM systems and ML infrastructure. Currently deep in multi-agent orchestration, code execution sandboxes, and replacing workflows that used to take days.

Himadri Mishra

I'm a generalist AI engineer with eight years across computer vision, NLP, MLOps, and agentic AI platforms. I've often been the sole or principal AI engineer at product companies, which means I ship the whole thing: model design, infrastructure, APIs, evaluation loops, and deployment. Most recently, I built an agentic platform at Knit that turned 48-72 hour analyst workflows into automated, consulting-grade research reports in under an hour, using LLM-generated code execution in sandboxes and a judge-LLM verification loop. I'm comfortable going deep on any layer of the stack, and I gravitate toward problems where AI can replace a real human bottleneck.

Knit
May 2025 – Apr 2026
Remote, India
Senior AI Engineer
Principal Architect, India AI Team
  • Designed and launched Knit 3.0, an agentic market research platform automating the full pipeline from raw survey data to verified insights, visualizations, and PPTX decks. Reduced post-fielding report turnaround from 48-72 hours to under 1 hour.
  • Built the insight execution engine: an LLM generates Python analytics code that runs in a persistent E2B sandbox reused across 30-50 tasks per report; a judge LLM independently verifies each result via its own sandbox, calibrated until AI output matched and surpassed the prior human-reviewed pipeline in accuracy, narrative quality, and visual output.
  • Implemented DAG-based task orchestration with topological sorting for maximum parallel execution; visualization pipeline generates 15-25 Vega-Lite charts per report with multi-threshold quality scoring.
  • Unified all agents onto a shared Python platform with multi-provider LLM routing (GPT-4/5, Claude Sonnet/Opus 4.x, Gemini 2.5 Pro), OpenTelemetry and Langfuse observability, pgvector RAG, SSE streaming, and auto-generated REST APIs.
Epic! for Kids
Feb 2023 – May 2024
Remote, India
Senior Research Engineer
ML Platform and Infrastructure Lead
  • Owned the full ML pipeline post-layoffs across discovery, recommendation, and search. Revamped ElasticSearch autocomplete, outperforming the prior solution in 80%+ of cases.
  • Cut Docker build time by 50%, reduced Kubernetes pod usage by 100x, slashed spot instance errors by 99%, bringing platform cost down by 10x.
  • Shipped ML features (book picker, performance-based carousels) via A/B testing; defined and executed ML roadmap resolving high-impact infrastructure issues.
Tangible Play (Osmo)
Sep 2019 – Feb 2023
Remote, India
Senior Research Engineer
Computer Vision Lead, India and US Teams
  • Set technical direction for CV across India and US; served as primary CV point of contact for the Worksheets product. Raised accuracy from 93% to 98%, impacting millions of learners.
  • Built a real-time U-Net shaded-region detection model achieving 80% IoU, deployed in Java on-device. Led tracing-the-dots games for Pre-KG to Grade 3, boosting engagement by 20%.
  • Automated tagging workflows in Go and Dart, cutting manual effort by 99%.
Whodat / Byju's
Jul 2018 – Aug 2019
Bangalore, India
Deep Learning Engineer
  • Built a fast ORB detector in C++ that was 20% faster than ORB-SLAM; researched monocular depth estimation (DeMoN, MvDepthNet). Team transitioned to Tangible Play / Osmo post-acquisition.
Generative AI
LLM Agents Agentic Workflows Multi-Agent Systems Prompt Engineering LLM Evaluation RAG AI Guardrails
Models and APIs
Claude Sonnet / Opus 4.x GPT-4 / GPT-5 Gemini 2.5 Pro Hermes OpenClaw Hugging Face
Frameworks
LangChain LangGraph LlamaIndex PyTorch TensorFlow Sentence Transformers
AI Coding Tools
Claude Code Codex Cursor Perplexity
Platform
FastAPI PostgreSQL / pgvector E2B Sandboxes ElasticSearch Redis Docker Kubernetes OpenTelemetry Langfuse Google Cloud
Languages
Python TypeScript Go C++ Java Dart
Rank 60 / 1120
FIDE & Google Efficient Chess AI Challenge (2025)
Top 6% globally in Kaggle's competitive chess AI challenge.
SN Bose Scholar
Research Intern, UC Berkeley (2017)
Trained neural programmer-interpreters using MuJoCo under Professor Dawn Song.
Internship
Software Engineering Intern, Microsoft (2016)
Built dialog-based chatbots for the Incentive Compensation team.
Competitions
Earlier Kaggle and Competitive Programming
Top 8% globally, Kaggle Allstate Claims Severity (2017). Honorary mention ACM ICPC Amritapuri Regional (2017). Honorable mention HP Think-A-Thon (2016).
Indian Institute of Technology (IIT-BHU), Varanasi
Integrated Dual Degree in Computer Science and Engineering
2013 – 2018
9.28
GPA / 10