Agentic AI Engineer · Lucknow, India

Building AI
that carries weight.

Faleha Qazi
FALEHA
QAZI
Agentic AI Engineer · Class of 2026
Available Now LangGraph Deployed ✓ RAG Pipelines FastAPI
scroll
LangGraph· FastAPI· Agentic RAG· Llama-3· Production AI· Eval Harnesses· Claude Code· FHIR R4 · HL7· Multi-Agent Systems· GCP · Docker · PyTorch·
// 01 — About

The person
behind the code.

I build agentic AI systems in regulated domains — healthcare, compliance, financial infrastructure. Places where a wrong output has real consequences.

My flagship project AutoAuth is a production-deployed prior-authorization agent with LangGraph, Llama-3, hybrid retrieval, UUID audit trails, and PII redaction. Built and shipped entirely alone.

I use Claude Code like a scalpel — to compress time, not skip understanding. I've collapsed two days of state-machine debugging into four hours. I also know exactly where it hallucinates. Knowing the difference is the skill.

Final-year B.Tech CSE (Health Informatics), VIT Bhopal. Graduating September 2026. Immediate joiner.

PythonLangGraph FastAPIClaude Code PyTorchDocker GCPFHIR R4PostgreSQL
StatusAvailable · Immediate
LocationLucknow, India
Open toRemote · Bangalore · NCR
EducationVIT Bhopal · 2026
Last RoleTech Mahindra Makers Lab
PublishedSpringer LNNS · ICAIC 2025
GitHubgithub.com/Falehaqazi
Emailqfaleha@gmail.com
Selected
Projects
01
AutoAuth-Clinical-Agent
Agentic RAG · LangGraph · Production · Render
End-to-end prior-authorization agent. Hybrid BM25 + semantic retrieval, reranking. UUID audit trail every node. PII redacted before retrieval. 14 commits, live. Caught 3 silent regressions with 40 adversarial test cases built from scratch.
LangGraphLlama-3FastAPIPostgreSQLStreamlit
FlagshipGitHub →
02
ClinTask
Multi-Agent · Google Gen AI Academy APAC Hackathon
Three-agent orchestration: router → retrieval → synthesis. Inter-agent protocol designed from scratch. Eval harness defined "correct" before building. GCP Cloud Run, Google ADK, Gemini 2.0 Flash.
Google ADKGemini 2.0GCPEval Harness
GitHub →
03
Clinical Triage RL
Reinforcement Learning · Meta PyTorch × Scaler Hackathon
Custom Gym environment for patient-triage RL. Reward shaping from clinical acuity. State encoding from raw vitals. Containerised, live on HuggingFace Spaces.
PyTorchOpenAI GymDocker
GitHub →
04
FHIRGuard
Clinical Validation · B.Tech Capstone
FHIR semantic validation layer, NEWS2 scoring, 200-bundle test fixture library. Full HL7 + US Core IG compliance checks. Modular pipeline for downstream ML.
FHIR R4HL7PythonUS Core IG
GitHub →
TECH
STACK
Python
LangGraph
FastAPI
RAG Pipelines
Claude Code
PyTorch
TensorFlow
Llama-3
HuggingFace
OpenAI API
Gemini API
Docker
GCP Cloud Run
AWS
PostgreSQL
Django REST
FHIR R4
HL7
FAISS
ChromaDB
WHERE I'VE
been.
Jan 2025
May 2025
Tech Mahindra · Makers Lab
AI Research Intern
  • Multi-modal Vision AI pipeline (Florence-2, PaliGemma, YOLO) on industrial drawings — VQA, detection, should-costing.
  • Designed agentic retrieval flow; owned annotation pipeline via Roboflow end-to-end.
  • Django REST backend (PostgreSQL, RBAC, notifications) — zero production incidents over 4 months.
2022
Sep 2026
VIT Bhopal University
B.Tech CSE — Health Informatics
  • Specialization: clinical data standards, medical AI, FHIR R4, HL7. CGPA 7.05.
  • Capstone: FHIRGuard. Published: ICAIC 2025, Springer LNNS.
  • Meta PyTorch × Scaler · Google Gen AI Academy APAC · GSoC 2026 Applicant.
// 04 — Published Research · Springer LNNS · ICAIC 2025
Early Detection of Sepsis from Clinical Data using Machine Learning Techniques
Jha A., Vaibhav K., Qazi F. et al.
Applied ensemble and classical ML to 40,000+ PhysioNet ICU records. Best model: Random Forest at 95.9% accuracy / 0.941 F1. SHAP interpretability surfaces clinically meaningful feature contributions.
95.9%
RF Accuracy
0.941
F1 Score
40K+
ICU Records
// 05 — Let's talk
LET'S
BUILD
something.
✉ Email me GitHub LinkedIn AutoAuth Live →
qfaleha@gmail.com