(03)Selected Work

A few things worth showing.

A cross-section of what I've shipped: security tooling, AI services, platforms, and a few experiments.

$ oravus scan ./ --agents 13
spawning 13 agents · 4 stages…
⚠ 3 critical · 2 chained paths
✓ consensus validated
2026

Oravus

Autonomous cybersecurity OS · Rust

13
security agents
10+
crate workspace
5
LLM providers

Rebuilt a single-model TypeScript scanner into a Rust multi-agent system. Designed 13 specialized security agents that collaborate over a private agent bus to discover, cross-validate, and chain vulnerabilities into attack paths. Runs offline-first with CI-ready output.

archRust workspace → discover → validate → simulate → report · agent bus + petgraph
RustTokioMulti-AgentAppSec
triage · safe
2026

Medical AI Triage

Telemedicine AI service · Freelance

42/42
adversarial safety evals
15
agentic tools
8B
fine-tuned LLM

Built the AI backend for a telemedicine platform. Designed dual-prompt clinical triage, fine-tuned and deployed an 8B medical LLM on RunPod (vLLM), added Whisper voice input, and shipped a 4-layer safety system with agentic tool-calling for appointments, prescriptions, and lab bookings.

archFastAPI → vLLM 8B + LoRA (RunPod) → 4-layer safety → Bedrock 15-tool agent
FastAPIvLLMLoRAAWS Bedrock
4 databases
2026

HerConnect

Workforce lifecycle intelligence

4
databases orchestrated
4
intelligent modules
ε=1.0
differential privacy

Built an AI platform for the women's workplace lifecycle. Implemented graph-based mentor matching, RAG leave-transition handovers, a privacy-first safety buddy system, and fairness-audited peer pods across four databases (Neo4j, Postgres, Milvus, Redis).

archNext.js 16 + FastAPI → Neo4j · Postgres · Milvus · Redis → GPT-4o (RAG + agents)
Next.js 16FastAPINeo4jMilvus
2026

Permory

LLM-maintained knowledge base

2D/3D
force-directed graph
187
automated tests
hybrid
vector + graph retrieval

Built a personal knowledge base that compiles notes, articles, and coding sessions into a persistent, interlinked wiki maintained by LLMs, so answers compound over time instead of re-running RAG on every query. Combines vector and graph-BFS retrieval over git-versioned markdown, with a 2D/3D mind-map for navigation.

archNext.js 14 → LLM ingest (journal + rollback) → ChromaDB + graph BFS → 3D force graph
Next.jsChromaDBThree.jsAnthropic
inairulesout
ziiro · 0 → 1
2025

Ziiro

AI job-application automation · 0→1

3
ATS platforms
96
automated tests
1-call
LLM form mapping

Built an end-to-end job-application agent that scans ATS portals (Greenhouse, Lever, Workday), ranks listings by fit, and auto-fills and submits applications via browser automation with single-call LLM form mapping. Wrapped it in a swipe-to-apply product on FastAPI with Redis queue workers.

archLangGraph + Playwright agent → ATS adapters → FastAPI · Postgres · Redis queue
LangGraphPlaywrightFastAPIRedis
on-prem
2025

Work Product Librarian

On-prem legal document AI · zero cloud

8
on-prem microservices
~18K
SALI legal concepts
BM25+FAISS
hybrid retrieval

Built on-premises document intelligence for law firms with zero cloud dependency. Ingests and OCRs legal work product, classifies it against the SALI taxonomy (~18K concepts), and powers hybrid BM25 and semantic search with LangGraph RAG Q&A, plus iManage and NetDocuments connectors.

arch8 FastAPI services (Nginx) → Postgres · Elasticsearch · FAISS → Celery NLP → React
FastAPIElasticsearchFAISSLangGraph
2025

Second Brain

Enterprise knowledge platform · Amdocs

50K+
files indexed
RAG
multi-agent flows
MCP
servers in prod

Built a multi-agent RAG system that processes 50K+ multi-format files. Designed LangGraph orchestration with vector databases and MCP servers for semantic search at scale.

archLangGraph multi-agent → vector retrieval → MCP tool servers
LangGraphMCPVector DBAgents
vector · search
2025

PeriodontalGPT

AI agent for dental health

1000s
patient queries served

Built a FastAPI backend with vector search for dental-disease diagnostics, serving thousands of patient queries. Implemented retrieval workflows, session tracking, and model-serving endpoints.

archFastAPI → Pinecone retrieval → HF model serving · session tracking
FastAPILangChainPineconeHugging Face
live · 97% auto-QA
2025

Convobrains

Automated call evaluation platform

−97%
manual QA effort
+18%
sales conversions
live
real-time dashboards

Built backend APIs, cron pipelines, and real-time dashboards that cut manual QA by 97% and lifted sales conversions by 18%.

archFastAPI → PostgreSQL/Redis → distributed cron → live dashboards
FastAPIPostgreSQLRedisCron
draft · IN
2024

Legal Lex

AI legal document drafting

Built a platform for drafting legal documents tailored to the Indian legal system, helping individuals and small businesses prepare filings with fewer errors.

archMERN → LangChain drafting flows → self-hosted Ollama inference
MERNMongoDBOllamaLangChain
AAPL · +2.43%
2024

FinsightAI

Real-time stock insights platform

Built full-stack microservices for high-frequency market data, with Redis caching on the hot path and AI inference modules for context-aware insights.

archmicroservices → Redis hot-path cache → AI inference modules
Node.jsFastAPIRedisRAG