Moving out of an apartment triggers a frustrating financial battle for many urban Indian renters. Landlords in cities like Bengaluru demand standard security deposits of ₹2 lakh to ₹3 lakh.
Yet, tenants struggle to recover this money. Landlords frequently invent arbitrary deductions for vague “wear and tear.” Tenants lack the legal clarity or resources to fight these unfair charges.
This systemic friction became a personal roadblock for product manager Siddharth Agrawal. Agrawal fought his landlord’s refusal to refund his deposit. His landlord fired back a dismissive taunt.
He claimed even the government eJagriti portal would fail. Agrawal refused to walk away. Instead, he spent his weekend engineering a technical solution.
He built Recourse. This AI-powered platform generates state-specific, filing-ready pre-litigation notices for tenants. The tool skips standard chatbot interfaces. It relies on a robust retrieval architecture instead. This approach directly tackles complex Indian tenancy disputes accurately.
The Fragmentation of Indian Tenancy Law
Building legal technology for the Indian real estate market is fundamentally difficult. The law itself remains geographically fragmented. Tenancy operates as a state subject. Therefore, rental rules change entirely across different jurisdictions.
Lawmakers introduced the Model Tenancy Act of 2021 to create a standard framework. It caps security deposits at two months’ rent. However, states adopt it voluntarily. Currently in mid-2026, only a few states actually follow this act. These include Tamil Nadu, Assam, and Uttar Pradesh.
Tech hubs like Bengaluru and Mumbai still use heavily localized, archaic laws. Here, landlords routinely demand up to ten months of rent upfront.
Complicating matters further, a vast majority of rental agreements are structured as 11-month “leave and license” contracts. This specific duration is intentionally chosen to bypass the Registration Act of 1908, which inadvertently weakens the document’s standing as hard evidence in court.
Technology lawyer Salman Waris points out that the core issue isn’t just tenant ignorance; it is the labyrinthine legal framework itself. For AI developers, this fragmentation is a serious technical hazard.
A standard, off-the-shelf language model prompted to write a legal notice will likely generate a convincing but legally disastrous document citing federal statutes where state laws apply or missing recent transitions like the shift from the Indian Penal Code to the Bharatiya Nyaya Sanhita.
Inside the RAG Architecture Powering Recourse
To mitigate legal hallucinations and ensure factual accuracy, Agrawal engineered Recourse around a highly localized Retrieval-Augmented Generation (RAG) pipeline. Rather than expecting an LLM to “know” the law, the system uses the LLM strictly as a drafting engine operating on retrieved, verified legal texts.
The user experience is heavily constrained to ensure high-quality outputs: tenants answer exactly seven targeted questions about their specific dispute. Behind the scenes, the architecture executes across three distinct layers.
First, the retrieval layer queries a vector database, currently operating with Redis, to isolate the exact state laws applicable to the user’s location.
For a Bengaluru dispute, the database forcefully retrieves clauses from the Karnataka Rent Act, the Consumer Protection Act 2019, Section 74 of the Indian Contract Act, and Section 80 of the CPC.
Next, an AI filtering layer evaluates this retrieved chunk of data. It strips out irrelevant statutes and legal noise, ensuring that only the laws directly pertaining to the tenant’s specific grievance are passed forward.
Finally, the generative layer takes over. Currently utilizing Gemini 2.5 while actively evaluating models like GPT-5.5 and Claude Sonnet for future iterations the LLM drafts the formal pre-litigation notice.
By anchoring the generative process entirely within state-specific legal boundaries, the platform successfully orchestrates a complex legal workflow into a highly accurate, accessible consumer tool.
Source: Official Analytics India Magazine, "How a 28-Year-Old Product Manager Built an AI Tool to Help Tenants Recover Rental Deposits"
