Arihant Deva is a developer and quantitative analyst building tools for research and coding workflows. Heimdall gives coding agents a searchable memory across repositories, with explicit checks on whether a retrieved path still matches its index.[1][4]
Start with Heimdall – Trust-verified knowledge layer for AI coding agents →
Education
Finance and Analytics studies at Rutgers Business School (in progress)
2025 – end date unverified
[1][2]
Links
Portfolio and professional chronology LinkedIn
GitHub · ArihantDeva 
X · DevaBuilds Senior Quantitative Analyst · Shepherd Ventures
Career
Deva’s owned chronology describes progressing from intern to senior quantitative analyst at Shepherd Ventures and earlier research work at FinBotX. His projects extend from finance into local agent tools. He records winding down Leviathan after pilots found no product-market fit. His newer Lumina wearable is still in development, with hardware in fabrication and its published specifications described as engineering targets.[1]
Contributions
Retrieve a memory and check its anchor
Heimdall combines locally computed semantic search, lexical matching and graph traversal across repositories. A local daemon watches files and updates a SQLite-backed index. Each result carries a status: its strongest card-based check compares file size and timestamps with indexed metadata, while other states identify weaker matches, moved files or dead paths. The metadata check does not reread the hit file or establish that a remembered claim remains semantically correct. It helps the agent decide which retrieved location to inspect before acting.[4]
Browse company career boards without submitting applications
Hunt supplies a self-hosted dashboard over discovered company career-board endpoints. Its server-side filters and search help a user narrow the collection by applicant-tracking system and recorded live status, then follow a board. Bookmarks and settings persist locally. The repository explicitly separates browsing from application automation, making it useful as an inspectable catalog rather than an unattended application workflow.[5]
Tools, writing & talks
Writing, research & talks
Start here
Local cross-repository memory with ranked retrieval and explicit path-verification states.
Start with the retrieval example and read what each verification state actually checks. You need the documented local runtime; inspect the returned file before relying on a remembered claim.
Creator and maintainer[4]Original source