Working Systems & Case Studies
Projects act as evidence of capability. Explore complete system architectures, data flows, and technical deep-dives across 9 production-grade applications.
RealEstate AI Sales OS
Stateful multi-agent workflow engine that orchestrates the real estate sales lifecycle through a directed graph with checkpoint-based persistence, MCP tool abstraction, and human-in-the-loop interrupts.
The core engineering challenge is orchestrating multiple LLM agents across a non-linear workflow with durable state. A lead doesn't move linearly from...
A production-grade multi-agent system demonstrating stateful workflow orchestration with checkpoint persistence, MCP tool abstraction for integration decoupling, LLM reliability engineering (structured outputs, model fallback, guardrails), and automated evaluation—reducing per-workflow LLM calls by 40% through agent consolidation and deterministic routing.
ThreadBase
LLM-powered content scheduling platform with async job orchestration, idempotent scheduling, and full audit trail observability.
Content teams struggle to maintain consistent social media presence across platforms. Manual scheduling is error-prone, lacks intelligent content gene...
A production-grade content scheduling platform achieving reliable job execution with complete lifecycle observability, efficient LLM cost management through semantic caching, and secure media delivery via signed URLs.
AI Compliance Audit Engine
Automating legal compliance audits for the Indian DPDP Act 2023 using a high-fidelity multi-agent AI system.
Legal teams spend significant time manually auditing privacy policies for DPDP Act compliance. Inaccuracies in these audits carry substantial regulato...
A regulator-ready audit system providing verified compliance tracking and high-fidelity reporting for digital privacy laws.
SpecOS
A development platform that synchronizes architectural specifications with production code through automated GitHub integration.
Software development often faces a disconnect between architectural design and implementation. Maintaining a consistent 'source of truth' becomes diff...
A development tool achieving synchronized repository states and low-latency architectural scaffolding.
Inbound Lead Engine
Production-grade async lead processing with AI intent inference and SLA enforcement.
High-volume lead ingestion creates bottlenecks. Needed a system resilient to AI failures and delivery issues without human intervention....
Full-stack containerized environment. Zero ingestion wait-time. Automated state escalation for hard failures.
Privacy-First Local RAG
An all-offline document intelligence system with OCR support and semantic retrieval for sensitive data sovereignty.
Sending sensitive data to cloud LLMs is a legal bottleneck for regulated industries (GDPR, HIPAA). Cloud processing is expensive at scale and introduc...
Zero cloud dependencies. 100% privacy compliance. Significant cost reduction by eliminating token-based pricing for high-volume internal document analysis.
Return Risk Predictor
Comparative study on predicting e-commerce returns using sentiment analysis and delivery metadata.
Wanted to predict 'Return' events before they happen. Returns aren't just bad products—they're the gap between expectation (reviews) and reality (deli...
Identified XGBoost as optimal for structured data. Peak AUC: 0.89.
MeetMe
Enterprise meeting intelligence with real-time transcription and local-first privacy-centric RAG.
Meetings are data-rich but forgetful. Standard tools are slow, expensive, raise privacy concerns by sending company assets to cloud for vectorization....
Low-latency platform with proactive context injection. No compromise on enterprise data sovereignty.
GST Compliance Copilot
Failure-aware RAG system for Indian GST compliance where 'I don't know' is a first-class feature.
In regulatory compliance, LLM hallucinations are legal liabilities. Standard RAG forces answers even with weak context. Unacceptable for GST tax laws....
Regulator-ready engine prioritizing factual grounding. 100% faithfulness in validation suites.
Forge — Coding Agent Runtime
Persistent runtime for reliable AI coding agents with controlled tool execution, workspace policies, persistent state, verification, and execution trajectories.
Coding agents need more than a model and a prompt. They need repository context, controlled access to tools, execution limits, persistent state, and a...
A working foundation for reliable coding-agent execution, with explicit boundaries around tools, state, verification, and long-running agent tasks.
Procurement Intelligence AI
Document intelligence pipeline for messy procurement data with parsing, OCR, structure-aware chunking, metadata enrichment, retrieval, and provenance.
Procurement documents arrive in inconsistent formats and contain structure that generic text extraction can easily destroy: sections, clauses, tables,...
A reproducible document-processing foundation designed for traceable retrieval over heterogeneous procurement documents.
Hackathon Intelligence Agent
Bounded, token-efficient agentic research system that plans targeted searches, executes cached research, compacts evidence, and synthesizes validated results.
Open-ended research agents can waste tokens, repeat searches, and pass large amounts of raw web content between model calls. The challenge was to keep...
A compact agentic research architecture designed around bounded tool use, evidence compaction, and deterministic post-processing.