I build AI systems for high-stakes decisions, where outputs need guardrails, evidence, and an audit trail—not just a convincing answer.
class AmanSaraswat:
name = "Aman Saraswat"
institute = "IIT Guwahati 🎓"
location = "India 🇮🇳"
focus = [
"Verifiable AI", "LLM Orchestration",
"FinTech AI", "Voice Interfaces"
]
languages = ["Python", "JavaScript", "SQL"]
currently = "Building AI systems with verifiable, grounded outputs"
open_to = [
"Collaborations", "Research", "SDE / ML Roles"
]
def mission(self):
return "Bridge the gap between powerful AI and regulated, high-stakes domains"- 🎙️ Built VoiceFinAI — a voice-first mutual fund advisor with a SHA-256 cryptographic anti-hallucination firewall, Deepgram Nova-3 + Groq LPU
- 🔒 Built FinShield RAG — RBI-compliant financial AI assistant with RAG, multi-layer security, and cryptographic audit trails
- 🐝 Built Swarm Ops — production multi-agent AI swarm for e-commerce with FastAPI + WebSockets + LLM orchestration
- 🤖 Explored AI Browser Automation with Playwright + AutoGen (AG2) + OpenAI
- 🫁 Applied deep learning to medical imaging (Pneumonia detection, 94%+ accuracy on chest X-rays)
- 💳 Built fraud detection models on real-world financial transaction data
| Project | Stack | What It Does |
|---|---|---|
| 🎙️ VoiceFinAI | Python Flask Groq Deepgram Edge-TTS |
Voice-first mutual-fund assistant with a SHA-256 AuditStream guardrail; Hindi/English |
| 🔒 FinShield RAG | Python RAG LLM FastAPI |
Financial AI assistant focused on retrieval verification, layered security, and auditability |
| 🐝 Swarm Ops | Python FastAPI WebSockets |
Multi-agent e-commerce system for real-time LLM task coordination |
| 🤖 AI Browser Agent | Playwright AutoGen OpenAI |
AI-driven browser automation using the AG2 multi-agent framework |
| 💳 Fraud Detector | Scikit-learn Pandas |
Proactive financial fraud detection — Random Forest + Gradient Boosting |
| 🫁 Pneumonia AI | PyTorch CNN OpenCV |
Deep CNN for pneumonia detection from chest X-ray images (94%+ accuracy) |
🤖 AI / ML / LLMs
⚙️ Backend & Infrastructure
📊 Data & Web
- Grounded answers — retrieval and verification before response generation
- Guardrails and traceability — deterministic checks and cryptographic audit trails for sensitive workflows
- Practical interfaces — fast, bilingual voice experiences designed for real users
"The best way to predict the future is to build it — with AI that you can actually trust."
⭐ Star my repos if you find them useful · Let's connect and build something remarkable
