📍 Open to AI Engineer roles at AI-first startups — Mumbai, Pune, Bangalore, Hyderabad, or remote.
I've built enough AI-integrated applications to know one thing for certain: the architecture meeting you skip always becomes the bug you can't find.
I design and build LLM-backed applications, multi-agent workflows, and retrieval systems with a strong emphasis on planning data flows and state boundaries before writing a single line of code. No black boxes, no hope-driven development—just predictable, maintainable backends that handle edge cases properly.
💸 FinOS | 🔗 Repository
Full-stack AI finance platform featuring an interactive dashboard and an integrated conversational engine sharing the same core backend logic.
- Hybrid Execution: Resolves roughly 60% of queries via pattern matching at zero LLM cost; fallback queries are routed to a Groq-backed orchestrator streamed over Server-Sent Events (SSE).
- Core Features: Financial health scoring, spending forecasting, and secure user data management built on top of a shared SQLite store.
🧠 hArI v2 | 🔗 Repository
RAG-based document intelligence system optimized for accurate retrieval and secure processing.
- Hybrid Retrieval: Combined dense vector similarity matching with keyword-based full-text search via custom SQL/RPC layers.
- Safe Sandbox execution: Replaced unsafe data frame executions with an isolated, in-memory DuckDB engine for structured data queries.
🤖 AI Agent Engine | 🔗 Repository
A modular 4-layer query routing and execution pipeline engineered to maximize reliability while minimizing external API dependence.
- Deterministic Evaluation: Implements an evaluation layer and localized semantic search ahead of external API calls, resolving up to 80% of repetitive queries efficiently.
- Cost Guardrails: Employs a strict Planner → Validator → Executor structure with hard session quota enforcement to prevent runaway execution loops.
🔬 Multi-Agent Research Pipeline | 🔗 Repository
A 6-agent supervisor-driven system designed for deep, stateful web research without context bleed.
- Stateful Boundaries: Segregates execution into explicit stages (Supervisor → Search → Scrape → Summarize → Critique → Synthesize) to ensure high data fidelity.
- Fallback & Controls: Features multi-provider fallback layers alongside Human-in-the-Loop (HITL) review gates to keep critical steps verifiable.
🎯 NextSteps | 🔗 Repository
An automated resume-to-JD gap analyzer that provides actionable skill-building roadmaps instead of arbitrary scores.
- Document Parsing: Extracts data from unstructured resumes and evaluates them against raw job descriptions or target URLs.
- Targeted Mapping: Directly correlates identified technical skill gaps to individual, step-by-step learning modules.
| Project | What it does | Stack |
|---|---|---|
| 🔌 DevMind — MCP Server | Local Model Context Protocol tool server providing LLMs with secure, human-gated filesystem access. | Python, MCP SDK, tiktoken |
| 🗄️ LangGraph Parallel SQL Runner | Executes parallel database queries utilizing graph state management, featuring inline review steps before query execution. | LangGraph, Groq, SQLite |
Open to AI Engineer roles — reach out