name: Farhaan Khan
role: Software Engineer
based_in: Hyderabad, India
degree: B.Tech CSE @ VIT Vellore (2026, CGPA 9.04)
currently:
- Building Reznt at Azntai, an AI agent paying businesses run on WhatsApp, Instagram and phone
- Shipping med-tech at Alfaleus: Droptor (smart eye-drop cap) and VisioVR
interests: [LLM agents, RAG, Full-stack, Android + hardware, Computer Vision]
Reznt ·    |
Multi-tenant AI revenue agent that sells, takes orders and recovers missed calls for paying businesses across WhatsApp, Instagram, Facebook and phone. Top contributor: 355 of 738 commits, 142 merged PRs. Durable workflows, Meta OAuth, Exotel voice, usage billing, hybrid search, LLM evals. (private repo) |
agent-eval-harness ·  |
Open source harness that runs 42 scripted support conversations against a fake shop backend and scores tool-calling agents on task success, policy, handoff, latency and cost. First run: gpt-oss-120b 97.6% at $0.0014 per scenario, Claude Haiku 4.5 90.5%, GPT-4.1 mini 83.3%. Recorded calls, so CI replays it with no API keys. |
Droptor ·   |
1st prize, Glaucoma Society of India Hackathon 2026. BLE smart-cap eye-drop adherence system: offline-first Flutter patient app plus a doctor dashboard with live updates. Live on the Play Store and App Store. (private repo) |
VisioVR ·   |
2nd prize, same hackathon. Built solo: Android app that simulates 12 eye conditions on the live camera in a phone VR headset with custom shaders at 60 to 120 fps. (private repo) |
Visor ·  |
VS Code extension that draws live flowcharts as you move through code, with complexity overlays and two-way line jumping. 1.4k+ installs, 10 merged PRs. |
rock-vs-docker ·   |
Same Flask API packaged three ways and scanned with Trivy: 1.12 GB and 3,609 CVEs with a basic Dockerfile, 137 MB and 179 hardened, 80 in a chiselled rock with zero from OS packages. |
Quantum Deepfake Detection ·   |
Research code for the IEEE Access paper below. |
A Quantum-Hybrid Framework for Enhanced Deepfake Detection
Integrating QAOA-Based Feature Selection with Quantum-Inspired Attention Mechanisms
IEEE Access, Vol. 14, pp. 17853 to 17873, 2026 · DOI: 10.1109/ACCESS.2026.3659021
Hybrid quantum-classical architecture for video deepfake detection: QAOA-based feature selection plus quantum-inspired attention. +2.5 points accuracy over classical baselines on FaceForensics++ and 45% lower calibration error.
