“Building production-grade legal intelligence, computer vision pipelines, and Linux kernel BLE systems — from architecture to deployment.”
The law governs us all, yet its language locks most people out. I build production-grade systems to break those locks.
Thelawgovernsusall,yetitslanguagelocksmostpeopleout.Advocateswastehoursonresearchthatshouldtakeminutes.Studentsstudyblind.Ibuildsystemstobreakthoselocks.Everyresponsegroundedinverifiedjudicialprecedent.Everycitationtraceable.Zerohallucinationenforcedatthearchitecturelevel.
NobleLex
AI
Next-Gen Legal Intelligence Platform
A production-grade Legal Tech platform engineered to revolutionize how Indian advocates, law students, and researchers interact with case law. Zero hallucinated citations.
Launch Platform> Query: "Anticipatory Bail under BNSS Section 482"
— Vectorizing query · pgvector(1536d) · 14,900 judgments
— BM25 keyword rerank · Cosine threshold 0.82
4 Verified Precedents Retrieved (142ms)
Sushila Aggarwal v. State (NCT)
Gurbaksh Singh Sibbia v. Punjab
NobleLex
Core Capabilities
Save 4+ Hours Per Case
Upload judgment PDFs — the system auto-extracts ratio decidendi, obiter dicta, key statutory sections, and the court's final holding in under 2 minutes.
Zero-Hallucination Citations
Every case reference retrieved from verified corpus before generation. If a precedent doesn't exist in the database, the system says so. Explicitly.
Court-Ready Drafts
Generate writ petitions, bail applications, affidavits pre-filled with correct statutory citations and jurisdictional formatting — ready for filing.
AI Court Simulator
Simulate oral argument sessions against an AI judge trained on Indian courtroom procedure.
Statutory RAG
Hybrid vector + keyword search across 14,900+ SC judgments. Zero hallucination at retrieval.
Criminal Law Bridge
Section-by-section mapping: IPC to BNS, CrPC to BNSS, IEA to BSA with annotated variations.
Drafting Studio
Generates jurisdiction-specific documents with verified statutory sections and court templates.
Cross
HID
Linux Kernel BLE Input Proxy
A high-performance daemon that turns any Linux machine into a universal Bluetooth LE HID proxy. Hooks into the Linux evdev subsystem for zero-latency input routing to iPads, macOS, and Windows — no client agents.
> evdev: /dev/input/event4 grabbed
— Registering BlueZ GATT Service: UUID 0x1812 (HID)
— LEAdvertisement1 registered on /org/bluez/hci0
> TARGET PAIRED: iPad Pro (Apple HID Descriptor Match)
CrossHID
Kernel & BLE Architecture
Zero Latency evdev
Captures raw hardware keystrokes directly from Linux evdev subsystem with sub-millisecond dispatch.
Agentless Compatibility
Target devices require zero software. Broadcasts as standard Bluetooth LE GATT peripheral.
Native Wayland Support
Bypasses display servers — operates on Wayland, X11, and headless Linux server TTYs.
Ultra Battery Efficient
Event-driven architecture idles at less than 1MB RAM and 0% CPU during non-active periods.
Antares
LM
Candidate Evaluation & Exam Intelligence
A specialized Ed-Tech AI backend for competitive exam preparation — covering NEET and mock judiciary exams. Dataset integrity, intelligent test assembly, and explanations that actually teach.
>> [CV] OMR grid detected... OK
>> [CV] Perspective skew corrected
Q1: A · 0.98 | Q2: C · 0.95
Q3: B · 0.99 | Q4: D · 0.96
>> [DB] Grading complete (0.38s)
AntaresLM
Core Architecture
Dynamic Dataset QC
Python ingestion pipeline validating question data against strict schema rules — flagging mismatches and duplicates.
RAG Explanations
Generates contextual explanations by retrieving the relevant formula, theorem, or legal provision per question.
Vision OMR Scanner
OpenCV pipeline processes smartphone photos of OMR sheets — detecting grids, identifying bubbles, grading instantly.
Algorithmic Assembly
Constructs balanced mock tests against defined spec: exam type, difficulty ratio, subject distribution.
From low-level Linux evdev D-Bus interrupts to 1536-dimensional semantic legal retrieval — every layer designed, verified, and deployed solo.
From Kernel to Cloud
Source Code Over Theory
I learn by reading real production codebases on GitHub, studying technical whitepapers from research labs, and reverse-engineering how mature systems actually work — not from passive video courses or textbook exercises.
End-to-End Ownership
I design, code, test, and deploy complete systems independently — from initial system design and database schema to the final pixel of the interface. No handoffs. No outsourcing. Total ownership of the outcome.
Relentless Iteration
Every difficult problem gets solved the same way: break it down, build the smallest failing case, fix it, and repeat. Late-night debugging sessions aren't a cost — they're how I actually understand how a system behaves under pressure.