CV
B.Tech in Computer Science & Engineering (Data Science) | AI Engineering • RAG • Interpretability • Applied ML
General Information
| Full Name | Hetansh Waghela |
| Location | Mumbai, India |
| hetanshwaghela06@gmail.com | |
| Phone | +91-8369107794 |
| GitHub | HetanshWaghela |
| hetanshwaghela |
Education
-
2024 - 2028 B.Tech in Computer Science and Engineering (Data Science)
Dwarkadas J. Sanghvi College of Engineering, Mumbai - CGPA 9.34 / 10
Experience
-
Mar 2026 - Jun 2026 AI Engineer Intern
AhumLabs, Mumbai - Built RAG for a voice/chat assistant on a multi-tenant trading platform, pulling tenant knowledge and past transaction history so users could fill bunker forms and look up vessel/pricing details in natural language.
- Shipped admin tooling for tenant onboarding, form customization, and knowledge uploads; each client's form fields and retrieval context feed directly into the assistant's tool schemas and prompts.
- Owned the embedding and indexing stack with local ONNX and Atlas Vector Search, including auto-indexing for new transactions and backfill scripts for existing data.
Technical Skills
-
Machine Learning
- Supervised & Unsupervised Learning
- Representation Learning
- Model Evaluation
- Error Analysis
-
Deep Learning
- Transformer Architectures
- Attention Mechanisms
- Fine-Tuning (LoRA)
- Probing & Interpretability
-
NLP
- Question Answering
- Retrieval-Augmented Generation (RAG)
- Text Classification
- Embeddings
-
Frameworks
- PyTorch
- Hugging Face Transformers
- LangGraph
- FastAPI
- Pydantic
- Instructor
-
Data & Tooling
- NumPy
- Pandas
- Neo4j
- ChromaDB
- Mem0
- Git
- Docker
Projects
-
2026 ANVIL - Compliance-Grade RAG Over ASME Standards
Python, Pydantic, NetworkX, sqlite-vec, FastAPI - Built a compliance-grade RAG system for pressure-vessel queries with auditable provenance, achieving 0.950 pass rate and 1.000 calculation correctness on a 100-example SPES-1 benchmark with NVIDIA NIM.
- Engineered a hybrid retrieval stack combining BM25, sqlite-vec embeddings, and a typed NetworkX knowledge graph via RRF and reranking, reaching 1.000 retrieval recall@10.
- Shipped a bounded agentic tool loop with transcript logging and guardrails; diagnosed loop failure modes and improved agent pass rate from 0.460 to 0.640.
-
2025 MythForge - Agentic Mythology Engine
Python, Gemini, LlamaIndex, Gradio, Neo4j, Mem0 - Architected a multi-phase generative pipeline transforming a single seed concept into a fully interconnected mythology using Gemini, LlamaIndex RAG, and Pydantic-validated entities.
- Built an MCP server with 7 tools enabling external AI agents to create, query, and expand mythologies programmatically.
- Engineered graph-backed memory and retrieval with Mem0 and Neo4j to maintain cross-entity consistency and long-horizon coherence across generated worlds.
-
2026 ImmunoPath - Immunotherapy Decision Support Pipeline
Python, PyTorch, LoRA, Hugging Face, Gradio - Built a 4-model clinical decision-support pipeline combining MedGemma, TxGemma, Path Foundation, and MedSigLIP to predict 8 immune biomarkers directly from routine H&E slides.
- Curated 60,396 patches across 950 TCGA NSCLC patients and diversity-selected 7,574 via K-Means for LoRA fine-tuning of a 4B vision-language model on A100-SXM4-80GB GPUs.
- Improved structured prediction reliability to 100% JSON/schema compliance and reduced TIL MAE to 0.16 on 94 held-out patients through response-only loss masking.
-
2025 Ghosts in the Model - Machine Unlearning & Interpretability
Python, PyTorch, Transformers, GPT-2, LoRA - Evaluated Gradient Ascent and NegLoRA unlearning on GPT-2 over 59 landmark-to-city forget prompts, reducing forget probability to 7.6% and 0.8%, respectively.
- Designed Lazarus activation patching to restore 89.7-100% of forgotten behavior, showing knowledge was suppressed rather than erased despite decodable residual representations in linear probes.
-
2026 ToolFusion - Middleware for Agent Tools
Python, Agent Tooling, Semantic Caching - Built middleware for AI agent tools with semantic caching, single-flight coalescing, deduplication, and cross-tool fusion.
- Designed the project around faster, cheaper, and more reliable agent execution in tool-heavy workflows.
-
2026 Multi-Domain GRPO Reasoning
JAX, Gemma, GRPO, TPU - Trained Gemma2 for transparent reasoning across multiple domains using GRPO.
- Reached 100% format accuracy and 60.8% exact accuracy in the Google Tunix Hackathon project.
Achievements
- 4th place (Creative, Track 2) at Anthropic & Gradio's MCP 1st Birthday Hackathon with 7,000+ global participants.
- Second Runner-Up at TechnoGrad 3.0 (DJSCE, Mumbai).
- Top 40 rank in DataHack 3.0, Mumbai's largest datathon.
- Open-source contributor to Stanford CRFM's HELM and TransformerLens.
Extracurriculars
-
Sept 2025 - May 2026 Co-Committee Member (Machine Learning)
DJS S4DS, Dwarkadas J. Sanghvi College of Engineering, Mumbai - Designed ML/NLP problem statements for hackathons and workshops; built dataset curation and automated evaluation pipelines.