ARIZONA STATE UNIVERSITY · RESEARCHJUN 2024 — PRESENT
AI Engineer — Public Health Assistant
- — Building an evidence-grounded health assistant with Dr. Mohan Tanirru — conversational guidance, biometric tracking, private report vault.
- — Fine-tuned open-source LLMs with LoRA/QLoRA (PEFT + Unsloth, 4-bit NF4) on clinical-dialogue data; served adapters on quantized GPU inference.
- — Own the eval pipeline: groundedness scoring, hallucination detection, regression suites gating every change, synthetic long-tail test generation.
- — Red-teamed against OWASP LLM Top 10 — prompt injection via uploads, jailbreaks — with automatic clinician-referral escalation. Live on AWS Amplify.
LoRA/QLoRAEvalsRed-TeamingAWS
MAAYA AI · STARTUPAPR — JUN 2026
Founding AI Engineer — Engineer #1 on AI
- — Owned the memory & retrieval layer end-to-end: ingestion, indexing, ranking and recall over meetings, emails, tasks and commitments.
- — Built the recall-quality eval harness: labelled retrieval sets, precision/recall-at-k, LLM-as-judge faithfulness scoring, regression gates per deploy.
- — Fine-tuned embedding & reranker models (LoRA) on in-domain pairs to beat off-the-shelf baselines — measured against the harness.
RetrievalMemoryLoRAGPU Inference
PROMACT · BHARATLAW.AI, VADODARAJAN — MAY 2026
Software Engineer Trainee — AI
- — Owned the LLM evaluation pipeline for a production legal-NER system: 4M-row ground-truth set, per-field scoring, failure taxonomy, CI-style regression gates.
- — Cut extraction cost 25x (₹1 → ~4 paisa/document) across an 8M-judgment corpus: 7-stage pipeline → 2-call Gemini-2.5-Flash-Lite, validated at 96% document-level zero-failure.
- — Eliminated hallucinated fields with DB/CSV-grounded structural extraction and windowed precedent extraction — groundedness as a measured metric.
- — Multi-agent legal systems: petition drafting, judge profiling, litigation strategy, lawyer recommendation.
EvalsGemini 2.5GroundingMulti-Agent
HIRAETH TECH · VADODARAJUN — DEC 2025
Full-Stack Developer — Intern
- — Shipped production services with NestJS, Next.js & PostgreSQL — API integrations and version-controlled deployments.
NestJSNext.jsPostgreSQL
PRODIGAL AI · REMOTEMAR — MAY 2025
Deep Learning Engineer — Intern
- — Fine-tuned transformers (HF Transformers, LoRA adapters) for NLP classification & extraction; contributed to LLM architecture for a finance client.
- — Trained CNN classifiers to 92% accuracy; built real-time YOLO/SSD detection pipelines.
PyTorchLoRAYOLOSSD
SPARKS TO IDEAS · REMOTEFEB — APR 2025
AI/ML Engineer — Intern
- — Trained an XGBoost churn model to 95% accuracy.
- — Built an LLM content-intelligence pipeline: FAISS HNSW multilingual dedup with an offline eval set for dedup precision.
XGBoostFAISS HNSWEvals
PAM INFOTECH · VADODARAMAY — JUL 2024
Summer Intern — Software Engineering
- — Built backend features in Java for internal web applications — the full request lifecycle from API to persistence.
JavaBackend
JEENWEB TECHNOLOGISTS · VADODARAMAY — JUL 2023
Web Application Developer — Intern
- — Client-facing web applications with React.js, HTML/CSS & JavaScript — responsive UI and API integration.
ReactJavaScript