Resume

Harshavardhan Manohar
857-308-7212 · m.h@northeastern.edu · linkedin.com/in/codevardhan · github.com/codevardhan · Boston, MA

AI Engineer (M.S. AI, Northeastern) who builds and ships production LLM/agent systems end-to-end. Built an LLM-driven agent from scratch alongside a founder - tool-calling orchestration, eval, guardrails, and deployment - cutting inference costs 88% at scale. Experience across the full AI lifecycle: data acquisition, model training, real-time pipeline design, and user-facing product. Background in processing noisy, unstructured real-world signals (medical imaging, continuous sign language video) where clean data doesn't exist and robustness is everything.

Education

Experience

Lead Machine Learning Engineer - June 2023 – August 2024

BeSuperhuman.ai · London, UK (Remote)

• Built an LLM-driven agent system from 0→1 with the founder: designed tool-calling orchestration that generates and executes browser actions (click/type/navigate) with guardrails, validation, and structured error handling.
• Developed the full eval and reliability stack: prompt + tool execution pipelines, structured logging, success/failure taxonomy, deterministic replay for debugging, and retry logic for production resilience.
• Reduced LLM inference and ops costs by 88% through caching layers, pipeline optimization, and SQL/stored procedure tuning - enabling high-volume runs without scaling spend linearly.
• Shipped cross-platform desktop app (Electron) and backend services (Python/Flask), with monitoring hooks and run analytics to close the loop between deployment telemetry and product iteration.

AI Research Assistant - May 2025 – August 2025

Massachusetts General Hospital / Harvard Medical School · Boston, MA

• Built an end-to-end deep learning pipeline for CT radiation dose estimation on ~12k DICOM scans - from raw data ingestion through preprocessing, model training, and structured evaluation reporting.
• Designed automated data acquisition and QC: ingestion from messy clinical sources, de-identification-safe preprocessing, normalization, and validation gates to catch distribution shifts and label issues before training.
• Achieved 12.3% error reduction vs. baseline through model iteration, experiment tracking, and reproducible evaluation - delivering structured results for non-technical clinical stakeholders.

Deep Learning Research Intern - June 2022 – January 2023

AmritaCREATE Labs · Amritapuri, India

• Developed a PyTorch transformer-based system for continuous sign language recognition from noisy, unstructured video - processing real-world signals where input quality varies wildly and clean data doesn't exist.
• Iterated on preprocessing and model architecture to improve robustness to low-fidelity inputs; communicated tradeoffs and experimental findings to non-technical stakeholders across e-governance teams.

Full Stack Developer Intern - January 2022 – April 2022

Agrisoft Dairy · Kollam, India

• Built mobile and backend features using Flutter and REST APIs; resolved integration issues across app and server components in a fast-moving small team.

Projects

• ParaGrad - From-scratch autograd engine & deep learning framework in C++17/CUDA. 19-op pluggable backend architecture, 886 tests, 500–1039× GPU speedup. Read more →
• KeepUp - LLM-powered relationship assistant with human-in-the-loop design: generates outreach suggestions, tracks user edits/rejections, feeds outcomes back into prompt and ranking iteration.
• Procedural Personalization via RL - RL-driven personalization loop (PPO) with telemetry-first design: per-second event logging, reward ablations, offline analysis pipelines for rapid iteration.
• Rocket Landing RL - DQN, PPO, and Safe PPO agents for 2D rocket landing with wind dynamics. Read more →
• EazyPredict - Python module for running & comparing multiple ML models at once with automated ensemble creation. Read more →

Technical Skills

• Languages: Python, TypeScript/JavaScript, SQL, Rust, C++, CUDA, Java, Dart
• AI/LLM: PyTorch, LLM agent orchestration (tool-calling, guardrails, retries), prompt engineering, eval frameworks, structured output parsing
• Data & Pipelines: ETL/data pipelines, feature engineering, EDA, dataset QC, SQL optimization; Pandas, NumPy
• Search & Retrieval: Embeddings, vector similarity search, ranking/recommendation signals, structured logging for RAG-style systems
• Infrastructure: Docker, CI/CD, experiment tracking, model monitoring, AWS (familiar)
• Backend & Apps: Flask, FastAPI, Django, Node.js/Express, REST APIs; Electron, Flutter