AI Engineer Co-op
Jan – Sep 2026
Staples Inc.
Built components of a production shopping assistant over a 200K+ product catalog. Worked on agent routing, fine-tuned models, graph retrieval, and a three-tier conversational memory system.
Harshit Brahmbhatt
Hey, I'm Harshit. I work on agents and AI systems, from memory and inference to CUDA kernels.
Most recently, I worked on conversational AI at Staples. I'm finishing my master's in Data Analytics Engineering at Northeastern, graduating in December 2026.
Selected work
Mnemosyne Agent memory Long-term memory for agents, from extraction to retrieval.
A five-stage async pipeline that turns conversations into structured memories. Built with PostgreSQL and pgvector, with full-text and vector search combined through reciprocal rank fusion.
Includes pluggable model providers, contradiction detection, and memory export and deletion APIs.
Read the sourceSpeech on Apple Silicon Speech recognition A speech model trained on a MacBook and deployed on the Neural Engine.
A 13.3M-parameter causal CTC head over a frozen EmbeddingGemma 2 audio encoder, trained on 960 hours of LibriSpeech using one M4 Pro MacBook.
The released Core ML model reaches 4.18% word error rate on test-clean and 10.71% on test-other. The streaming build has 207 ms median word latency. Evaluated on English read speech.
Read the model cardNanbeige on MLX Model porting A custom Apple Silicon port of a recurrent-depth language model.
Wrote the MLX architecture module for Nanbeige4.2-3B, including its repeated decoder passes and separate KV caches, then converted it to 8-bit weights.
Validated against the official Hugging Face implementation: all 40 greedy output tokens matched in the reference comparison.
Read the model cardExperience
ML Engineer Intern
Jan – Jul 2024
Pixeltechnologies
Built ad-engagement prediction pipelines with scikit-learn and XGBoost, and SQL and Airflow workflows for experiment data and model retraining.
Data Analyst Intern
Jul 2023 – Jan 2024
GROWITUP
Built semantic search across 5K+ documents using OpenAI embeddings and FastAPI. Analyzed user sessions with Snowflake and PostgreSQL.
Education
Northeastern University
2024 – 2026
MS, Data Analytics Engineering
Boston · Expected December 2026
LDRP Institute of Technology
and Research
2020 – 2024
BE, Computer Engineering
Gandhinagar, India
What I work with
- Languages
- Python, C++, SQL, TypeScript
- Models & agents
- PyTorch, Core ML, MLX, vLLM, LangGraph
- Data & retrieval
- PostgreSQL, pgvector, Neo4j
- Shipping
- FastAPI, Docker, MLflow, Databricks, AWS
Get in touch
For work, a project, or a conversation about what you're building:
brahmbhatt.h@northeastern.edu