About • Currently Building • Featured Builds • Hardware Engineering • Tech Stack • GitHub Analytics • Achievements • FAQ • Open To Collaborate

$ whoami
Aneesh Venkatesha Rao — B.Tech ECE, NIT Warangal (2024–2028)

$ status
AI Engineer • Open Source Contributor • SIH 2025 National Finalist (Top 5 of 75,000+ submissions)

$ focus --areas
- AI/ML product engineering (RAG, LLM orchestration, retrieval systems)
- Full-stack application architecture
- Distributed data + automation pipelines
- Human-in-the-loop developer tooling

$ currently
Porting Uktam.ai to iOS, building RiskMesh for the Razorpay AI Buildathon,
and working toward open-source contributor credit (LFX Mentorship / GSoC).

I enjoy bridging research-minded experimentation with production-ready software. My approach combines AI/ML capability, backend rigor, and clean frontend execution to ship products that are both technically strong and genuinely usable.

Uktam.ai — Offline Indic speech-to-speech translation

Co-founded app for offline, on-device speech translation across Hindi, Kannada, Tamil, and Telugu. Android app is live; currently leading the iOS port (SwiftUI) alongside website ownership and Android QA/testing.

Tech: Sherpa-ONNX, llama.cpp, Sarvam Translate, SwiftUI

🔗 Website: uktam-web.vercel.app · Google Play: Uktam.ai

ContextCraft · AST-aware codebase search engine  

• Built a hybrid search engine that parses codebases with tree-sitter — functions and classes as atomic chunks, never mid-function splits
• Implemented Reciprocal Rank Fusion merging pgvector cosine similarity + PostgreSQL BM25, with per-repo RRF normalization for multi-repo queries
• Added Cohere cross-encoder reranking (20 → 10 candidates) and 1-hop dependency graph expansion with cycle detection across file boundaries
• Enriched every chunk with git blame author metadata; benchmarked at 80% source hit rate at 3.88s P50 latency across 30 queries
• Published to PyPI (pip install contextcraft-py); CI enforces mypy strict, ruff, and pytest before every merge
• Tech: Python, FastAPI, PostgreSQL, pgvector, tree-sitter, Cohere, Next.js, Docker

🔗 Repository: github.com/AneeshVRao/ContextCraft

RiskMesh · Coordinated payment abuse-ring detection  

• Built for the Razorpay AI Buildathon (Track 02: AI Risk Manager) — graph/relationship-based detection of coordinated fraud rings, not just single bad actors
• Designed a tiered architecture: deterministic baseline → XGBoost → GNN stretch goal, with ring-level leakage-safe evaluation and a false-positive cost model
• Shipped Tier 0 end-to-end: config, synthetic ring/hard-negative generators, graph hygiene checks, deterministic scorer, chronological ring-level split, integrity reporting, and a full evaluation runner
• Caught and resolved two real signal bugs pre-ML: an inverted ip_concentration direction and a flawed temporal_burst signal, both fixed and tagged before building on top of them
• Tech: Python, XGBoost, graph-based feature engineering

🔗 Repository: github.com/AneeshVRao/RiskMesh

Stratos · "Football, Understood" — tactical AI explainer  

• Built for the IBM AI Builders Challenge — spans two tracks: a football tactical decision explainer and a multilingual fan chat companion
• Built on IBM Granite / WatsonX, Docling, and Langflow, with a Context Forge-backed retrieval layer
• Frontend built in D3.js for tactical visualizations, backed by a stateless, per-call backend
• Tech: IBM Granite, WatsonX, Docling, Langflow, Context Forge, D3.js

🔗 Repository: github.com/ashb155/fifa-ibm _{(team repo, hosted under collaborator's account)}

Humanify · AI text humanizer — live, monetized SaaS  

• Built a production SaaS that rewrites AI-generated text to read naturally, with multi-AI routing across Gemini and Claude depending on input characteristics
• Shipped and currently runs a paid Pro tier (₹999/mo) with Razorpay billing integration
• Designed the full subscription lifecycle — auth, billing, plan gating — on Supabase
• Tech: Next.js, Supabase, Razorpay, Gemini API, Claude API

🔗 Repository: github.com/AneeshVRao/Humanify

Dev-Saarathi · Voice-first AI coding assistant  

• Built a voice-to-code pipeline in 11 Indian languages through transcription + intent detection
• Designed an AI orchestration flow (Intent Detection → Guardrails → Execution Router) for safe automation of code actions
• Implemented context-aware reasoning over 50+ files / 100k+ characters via RAG to reduce hallucinated outputs
• Added human approval controls while keeping 3–6s voice-to-response latency
• Tech: TypeScript, Python, AWS (Bedrock, Transcribe, S3), RAG

🔗 Repository: github.com/ashb155/dev-saarathi _{(team repo, hosted under collaborator's account)}

ShabdSetu · AI-powered multilingual learning platform  

• Built an end-to-end learning system for 10+ Indian languages using IndicTrans2 + Whisper
• Developed fuzzy pronunciation scoring for real-time speaking feedback
• Engineered layered caching (in-memory + Firestore) for sub-50ms response paths
• Added gamification with XP, streaks, and 27 achievements synced across authenticated sessions
• Tech: Next.js, React, FastAPI, IndicTrans2, Whisper, Firestore

🔗 Repository: github.com/AneeshVRao/ShabdSetu

Nexus Load Balancer · HTTP load balancer in Go  

• Built a lightweight HTTP load balancer from scratch with round-robin request distribution across backend pools
• Implemented active health checks (periodic probing) and passive health checks (live-traffic failure detection) to pull unhealthy backends out of rotation automatically
• Tech: Go, net/http

🔗 Repository: github.com/AneeshVRao/Nexus

Coursework and lab projects at NIT Warangal, several built alongside classmates Akula Sahasra, Adhvay Shrujal, and Arushi Pundir.

Bayesian In-Memory Compute (IMC) Core SRAM-based weight lookup, LFSR-driven stochastic sampling, a Kogge-Stone parallel-prefix popcount unit, and FSM control logic — implemented and verified in Vivado.

FPGA Brain Tumor Segmentation Otsu thresholding + watershed segmentation accelerated via Vitis HLS on a MicroBlaze soft core, achieving a 144–229× speedup over a software baseline. Deployed on Nexys 4 DDR / Artix-7.

Tech: Verilog, Vivado, Vitis HLS, MicroBlaze

• Ship, don't just prototype — Humanify runs live billing, Uktam.ai is on the Play Store; production behavior over demo behavior
• Measure before claiming — ContextCraft's 80% hit rate and Stratos's speedup numbers come from actual benchmarks, not estimates
• Catch bugs before they compound — found and fixed two inverted scoring signals in RiskMesh before building ML on top of them
• Build for underserved languages — Uktam.ai and ShabdSetu both target Indic-language users most AI tooling ignores
• Keep retrieval and reasoning systems explainable — RAG pipelines and rerankers should be debuggable, not black boxes

• 🥇 Smart India Hackathon 2025 — National Finalist (Top 5 in problem statement, selected from 75,000+ submissions)
• 📜 13 professional certifications across cloud, software engineering, and data science (Google, Meta, IBM)
• 🏅 65+ Olympiad medals in national-level Mathematics and Science competitions

Who is Aneesh Venkatesha Rao? Aneesh Venkatesha Rao is an AI engineer and full-stack developer, and a third-year B.Tech Electronics & Communication Engineering student at NIT Warangal, India (Class of 2028). He is a Smart India Hackathon 2025 National Finalist.

What has Aneesh built? Notable solo builds include ContextCraft (an AST-aware code search engine combining tree-sitter parsing, hybrid RRF search over pgvector and PostgreSQL BM25, and Cohere reranking, published to PyPI), RiskMesh (a graph-based fraud-ring detection system built for the Razorpay AI Buildathon), and Humanify (a live, monetized SaaS that rewrites AI-generated text using multi-provider LLM routing). He co-founded Uktam.ai, an offline on-device speech translation app for Indic languages, live on Android with an iOS port in progress. He has also contributed to team projects including Stratos (a football tactical AI explainer for the IBM AI Builders Challenge), Dev-Saarathi (a voice-first AI coding assistant), and ShabdSetu (a multilingual language-learning platform).

Does Aneesh have any live, user-facing products? Yes — two. Humanify is a monetized SaaS with a paid subscription tier and live billing. Uktam.ai is a published Android app on the Google Play Store, with an iOS port underway.

What technologies does Aneesh work with? Python, TypeScript, Go, C++, SQL, FastAPI, Next.js/React, PostgreSQL, pgvector, Docker, and LLM/RAG tooling including LangChain, Cohere, Whisper, and IBM Granite/WatsonX. He also has hardware engineering experience in Verilog and FPGA development (Vivado, Vitis HLS).

Is Aneesh open to internships or collaboration? Yes — he is actively looking for AI/ML product engineering internships, backend/full-stack roles, and open-source collaboration (targeting LFX Mentorship and Google Summer of Code contributor credit). Reach out via email or LinkedIn.

• 🌱 Open-source contributions — actively working toward LFX Mentorship / Google Summer of Code
• 🤖 AI/ML product engineering internships
• 🛠️ Backend / full-stack engineering roles
• 🔗 Collaborations in dev tooling, RAG systems, and developer productivity

If you're building something meaningful in AI, full-stack systems, or developer tooling, I'd love to connect.