Heer Patel
AI Researcher & Engineer
Powered by chai, curiosity, and the belief that technology should feel human before it feels intelligent.
About
Who I Am
I have always been drawn to the space where AI meets real people with real problems. Not the benchmarks, not the papers, but the moment someone's life gets a little easier because a system actually worked for them. That is what drives everything I build.
🎓I recently finished my Master's in AI at Boston University, working across research, engineering, and leadership, sometimes all at once. It has been the most humbling and rewarding chapter of my life so far.
🔬As a GenAI Researcher at BU's BIT Lab, I worked alongside The Washington Post to understand what really happens after a generative AI gives you an answer. I also reviewed Context Engineering for Multi-Agent Systems for Packt. Both experiences reinforced the same belief: the hardest problems in AI are not technical. They are about trust between systems and the people who use them.
🛠️On the engineering side, I have built and shipped things across internships, production environments, and client-facing products. Leading a team as a Technical PM taught me something that has stuck with me: great engineering is really just great listening.
👩🏻🏫Teaching is where it all clicks together. As a Graduate Teaching Fellow, I supported students in building multimodal AI systems from the ground up. Watching someone go from complete confusion to a working model is honestly one of the best feelings this field has to offer.
⚠️Side effect of hiring me: unreasonably high output.
Experience
Where I've Worked
Generative AI Researcher
Questrom School of Business, Boston University · Boston, MA
Research under Prof. Dokyun Lee
- Partnered with The Washington Post to investigate user behavior in generative AI search — uncovering how query intent shapes engagement and personalized search design.
- Trained BERTopic on 120k+ article headlines to construct a shared topic space across queries, AI-generated answers, and article content.
- Quantified semantic drift using Jensen-Shannon distance across query-answer-source pairs, enabling formal analysis of generative search alignment.
Technical Reviewer
Packt · Remote
Reviewed “Context Engineering for Multi-Agent Systems” by Denis Rothman for technical accuracy and clarity
ML Technical Project Manager
BU Spark! · Boston, MA
- Oversaw end-to-end lifecycle of 4 external client ML products; optimized project velocity by 30% via Scrum boards.
- Reduced debugging cycles by 25% through rigorous code review and architectural mentorship for 15+ developers.
Graduate Teaching Fellow
Boston University · Boston, MA
- Guided 60+ students through PyTorch pipelines for neural modeling — increasing project completion rates by 40%.
- Architected template pipelines for multimodal tasks; student teams achieved ~90% accuracy in emotion recognition and symptom detection.
Machine Learning Engineer
The AI Collective · Boston, MA
- Engineered a constraint-aware semantic matching engine with community governance rules, improving match precision by 25%.
- Optimized vector search pipeline, cutting batch runtime by 50% and achieving 85% SLA compliance under high load.
AI Engineering Intern
SAP · Gujarat, India
Selected among top 60 students across Gujarat
- Built an AI recruitment platform using NLP-based resume-JD matching and automated soft-skill interviews — 87% classification accuracy.
- Implemented extraction, semantic mapping, and sentiment analysis for talent-role alignment.
AI Intern
RadicalX · Remote
- Used OpenAI, VertexAI, and TensorFlow to enhance ReX, an AI coaching product.
- Built personalized career guidance and mentorship features serving 500+ learners.
Data Science Intern
BrainyBeam Technologies · Ahmedabad, India
- Built ML models to predict bank marketing campaign performance; optimized SQL queries reducing retrieval time by 10%.
- Aggregated 1k+ real-time insights from operational data for SalesForce case investigation.
Machine Learning Intern
Cygnet Infotech · Ahmedabad, India
- Created custom ML applications for critical predictions, automated reasoning, and optimization algorithms.
- Prototyped ML applications and evaluated application vulnerability early in development.
Projects
Things I've Built
Vagueness Analysis of VLN Models
Jan - May 2025
Built a dual-metric framework (lexical + LLM-based) to quantify instruction vagueness across VLN datasets. Analyzed effects on SoTA VLN agent generalizability and explained architectural robustness factors.
UnMask: Occlusion-Resilient Face Recognition
Jan - May 2025
Improved occluded-face recognition by 25% using a frequency-aware Siamese network with LBP-based alignment. Achieved 36.4% partial and 21.9% masked similarity gains over baseline.
Real-time Lane Traffic Density Detection
Jan - May 2025
Trained a Deep Q-Network agent for adaptive traffic signal control. Reduced estimated vehicle travel latency by ~20% over static baselines across simulated multi-lane intersections.
Sequential Sentiment on Video Transcripts
Sep - Dec 2024
Fine-tuned DistilRoBERTa for minute-level sentiment classification on long-form video. Built temporal affective trend visualizations to surface narrative shifts and improve content interpretability.
Skills
Tech Stack
Languages
ML & Deep Learning
AI Specializations
Tools & Infrastructure
Education
Academic Background
M.S. Artificial Intelligence
Boston University
Boston, MA
B.E. Computer Science
Gujarat Technological University
Ahmedabad, India
Contact
Let's Connect
Open to AI/ML roles, research collaborations, and interesting conversations.
Say Hello
Whether it's a role, a research collab, or just a chat about AI — my inbox is always open.
Send a Message