Home Interactive Lab ⚡ Projects Spaces Hub Writing About Skills Experience Contact
Open to new opportunities

Hi, I'm Nirmit K. Tripathii AI / ML Engineer & Researcher

Building production-grade AI systems — multi-agent LLM orchestration, multimodal RAG, speech & video ML, legal AI, and civic technology. 3+ years of hands-on ML engineering, plus open-source and applied research.

Launch Interactive AI Lab View Projects Live Spaces Hub Download CV
3+
Years Experience
13+
AI Projects Shipped
22
Indic Languages
11
Deep-Dive Write-ups
What I've built

Featured Projects

Flagship production systems, research implementations, and agentic AI architectures.

EN Audio
6.50s
HI v2 Raw
11.80s
HI v3 QLoRA
6.48s
Length Error: 0.103 rel (79% cut) · Syllables Synced
Flagship Case Study
Llama-3.1-8B QLoRAIndicF5DemucsIsochrony

Indic AI Dubbing Platform

Isochrony-aware video dubbing pipeline translating English to 11 Indic languages with zero-shot voice cloning. The v3 fine-tune constrains translation to a strict phoneme budget, cutting length error 79% (0.495→0.103 rel.) while improving translation quality (chrF++ 22.0→30.1).

Build Log → Walkthrough Repo
mcp-email-daemon · stdio
> mcp_server.fanout_query("action_required: true")
↳ [Gmail] 200 OK (3 threads) · [Zoho] 200 OK (1 thread)
↳ [Outlook] SIGTERM detected → self-healed in 42ms
✔ 17 unified tools aggregated · 49 tests passing
Flagship Case Study
MCPMulti-AgentTypeScript17 Tools

Email Orchestrator — Multi-Account MCP Agent

A multi-account email agent consolidating Gmail, Zoho, Yahoo & Outlook into one AI-enriched inbox — a single MCP server that is itself an MCP client, fanning out to each provider and re-exposing 17 unified tools with self-healing child process reconnection. 49 passing tests.

Build Log → Resilience Log Thinking Budget
India Aaspaas Noida / Sec 62 / Sec 59
Open pothole · Sector 62
Garbage disposal · Sector 59
Water & Drainage · Mamura
Live OpenStreetMap
Full-StackNode/PostgresCivic TechSSR

India आसपास — Civic Reporting

Production civic-issue reporting platform: residents map potholes, garbage and drainage across Noida and NCR, with server-rendered social share cards, a two-tier trust model, and a hardened upload pipeline — all running for about $3/month.

Live App Repo White Paper
gitscout · trust-engine · prefilter_farm
> scrape("algora.io in:body") → 41 raw hits
↳ blocklist: bounty-plaza rejected (crypto-deposit scam)
↳ engagement gate: rustchain-bounties 20 → 0 admitted
✔ 18 trusted candidates · 11/11 trust tests green
FastAPI (async)Next.jsTrust EngineAnti-Fraud

GitScout — Trust-Filtered Bounty Intelligence

An open-source bounty & issue aggregator whose moat is telling the truth about a bounty — including when the truth is "don't bother." A pure-function, three-layer trust engine rejects engagement farms and crypto-deposit scams before any network call; live-validated when a real farm's admitted count dropped from 20 to 0.

Live App Field Report White Paper Series
01. Ingest
04. Hook
10. QA Gate
Fallback: Groq ➔ Gemini ➔ Ollama ● QA Passed
Agentic AIMulti-AgentOpen SourceOrchestration

Podcast Shorts Factory

Ten cooperating AI agents that turn long podcasts into scheduled short-form videos — a provider-agnostic LLM layer with automatic fallback, a learning feedback loop, and a QA gate. Runs unattended on free providers. MIT-licensed.

View Repo → White Paper
Ingest: youtube.com/@veritasium 148 WPM · Hooked
1. Multi-LLM Cascade Groq (429) ➔ Gemini Flash (200 OK)
Failover OK
2. Self-Healing JSON Schema Repaired token AST in 18ms
Healed
3. Neural Audio & MoviePy Stack Edge-TTS ➔ 1080p MP4
$0 / Free Tier
StreamlitLLM FailoverSelf-HealingZero-Cost

AI Video Style Cloner

Paste a YouTube channel URL and it reverse-engineers the style, writes an original script in that voice, narrates and illustrates it, and compiles a finished MP4 — for $0. A three-tier LLM failover, self-healing JSON validators, and a zero-network procedural generator guarantee output even when every API is down.

White Paper →
"अग्निमीळे पुरोहितं..." (Rigveda 1.1.1) 0.942 Cosine
Knowledge Star Map 1,500 / 14,205 Nodes
Bhaskara · Siddhanta-Siromani (Calculus)
Sanskrit-MiniLM · LightRAG Vedic GraphRAG
RAGLangGraphSanskrit NLPIKS

IKS Oracle — Gyanstrot

AI-powered Indian Knowledge System oracle enabling semantic querying of Vedic texts, Upanishads, and ancient Indian manuscripts using multilingual RAG & graph-based retrieval.

View Repo → Live Demo
Bengaluru vs Chennaiyin (11') +0.742 Net VAEP
9m24s Pass · Udanta (+0.024)
10m19s Corner · Roshan (+0.088)
11m35s ⚽ Goal · Basafa (+0.785)
V(a) = ΔP(scores) − ΔP(concedes) StatsBomb Live
Sports AIPythonStreamlitWorld Cup

VAEP Football Action Valuation

Interactive dashboard for valuing football actions using the VAEP framework. Analyzes match events across Top 5 European Leagues and World Cup 2018 to rank players and visualize goal-scoring sequences.

View Repo → Live App
Coordinates: (88.5, 50.0) Shot Angle: 36.8°
0.785 xG
● Goal Probability: High
MLXGBoostHeatmap

Football Analytics — xG Model

Interactive Expected Goals (xG) simulator with pitch-mapping and heatmap visualization. Evaluates scoring probability in real-time.

View Repo → Try Simulator
SBERT all-MiniLM-L6-v2 91.4% Cosine Match
91.4%
✔ LangGraph 98% ✔ PyTorch 94%
✔ Indic NLP 91% ✖ Gap: Rust
Dense + TF-IDF Reranker Tier-1 Skill Match
NLPSBERTSkill Gap

AI Resume Matcher

Advanced semantic matching engine using Sentence-Transformers. Features skill extraction and gap analysis for job applications.

View Repo → Analyze Resume
Lesion [88.4%]
Solanum lycopersicum (Tomato)
Tomato Early Blight
98.4% Alternaria solani
1.2% Late Blight
Rx: Copper Oxychloride 50 WP (2.5g/L)
Deep LearningPyTorchComputer Vision

Plant Disease Classification

CNN-based mobile diagnostic tool for identifying crop diseases in real-time. Achieved 98% accuracy on the PlantVillage dataset.

View Repo → Live Demo
OpenWeatherProxy API Global Doppler
28°C Partly Cloudy
💨 14 km/h NW 💧 62% Humidity 🧭 1013 hPa
Cache HIT · 12ms 97% Bandwidth Saved
APIProxyThrottled

Global Weather Dashboard

Real-time weather tracking with a secure proxy layer. Implements global rate limiting and intelligent caching to manage API quotas.

View Repo → Check Weather
Interactive AI Demos

Live Hugging Face Spaces Hub

Explore and test live production ML models deployed on Hugging Face Spaces. Profile: huggingface.co/nirmitk

Live Space

IKS Oracle — Gyanstrot

AI-powered Indian Knowledge System enabling semantic search & graph-based exploration over Vedic texts and ancient Indian manuscripts with multilingual embeddings.

Launch App Space Code
Live Space

VAEP Football Action Analyzer

Action-valuation dashboard quantifying offensive and defensive match events across Top 5 European Leagues and World Cup 2018 using the VAEP framework.

Launch App View Repo
Live API

Shot Quality (xG) Model

Interactive pitch coordinate mapper and statistical scoring probability estimator powered by a Random Forest Classifier trained on European league shot events.

Pitch Simulator Model Code
Live API

AI Resume Compatibility Matcher

Deep semantic matching and skill gap analyzer comparing candidate resumes against job descriptions using Sentence-Transformers (SBERT all-MiniLM-L6-v2).

Match Resume API Space
Live API

Plant Disease Classifier

Deep CNN vision diagnostics identifying 38 crop foliar diseases in real-time from leaf imagery with 98% validation accuracy on the PlantVillage dataset.

Test Leaf Image View Repo
Profile Hub

All Spaces & Models

View the complete catalog of open-source datasets, model weights, and live web applications hosted under the @nirmitk Hugging Face organization.

Open Hugging Face
What I write

White Papers & Blogs

Deep-dives on the engineering and research behind my projects — written to be read, not skimmed. From production systems to primary-source history.

Field notes — straight from the repo
Isochrony Fine-Tune 49% → 10% Err ● Llama-3.1-8B QLoRA
Fine-TuningQLoRAIsochrony

Teaching an LLM to Fit Its Mouth

AI dubbing drifts because translations aren't the same length as the original. The industry fixes it in the audio; I fixed it in the translator — cutting length error from 49% to 10% on a fine-tuned Llama-3.1-8B, with the full training telemetry left in.

202612 min read →
Kaggle MLOps T4 GPU · 15.2 GB ● Reverse Tunnel Out
KaggleGPUTunnel

A Headless GPU Lab on Kaggle's Free Tier

A GPU I didn't pay for, running an app I could open from my phone, that I could kill from anywhere — a tunnel out, a public message channel back, and a kill switch. Plus the meta-tensor bug that cost the most time.

20269 min read →
SoccerNet 2026 mAP 17.69 (Avg) ● Decomposed Anticipation
AnticipationSoccerNetTransformers

Predicting What Happens Next in Football Video

Detection asks what just happened; anticipation asks what's about to. My SoccerNet 2026 scores say the model knows what is coming far better than when — and the decomposed metric shows exactly which half broke.

202610 min read →
The Email Orchestrator · a 3-part build log
Part 1 · MCP Topology 1 Server ➔ 4 Inboxes ● 17 Unified Tools
Fan-out17 tools

One Inbox, Three Mailboxes, Zero New APIs

The architecture: one MCP server that is also an MCP client, consolidating four mail providers behind a single 17-tool surface — with exactly one custom component.

202612 min read →
Part 2 · Reconnection allSettled Isolation ● 49 Passing Tests
ReconnectionallSettled

One Dead Inbox Shouldn't Kill the Whole Query

An MCP connection is a child process, and child processes die. The self-healing reconnection — isolation, drop-detection, and a cooldown — that makes it survivable, with two production bugs pinned in tests.

202611 min read →
Part 3 · Token Economics Floor Budgeting ● Thinking Saturation
Thinking ModelsToken Budget

The Model Spent Its Whole Budget Thinking

Empty JSON from an 80-token task with a 512 budget. On a thinking model the fix for a truncated answer is a bigger budget — and it fails completely silently. Floor it, double on retry.

202611 min read →
GitScout · trust-filtered bounty intelligence — a field report + 3-part series
Field Report · Strategy H4 Falsified ● Honest Negative Result
Anti-FraudLive Verification

The Bounty That Told Me Not to Bother

I verified four real Algora bounties live and found every winnable-looking one saturated or already paid. The honest move was to report the negative result, not manufacture a win — which is exactly what the product promises its users.

202610 min read →
Part 1 · Research ~40% Farm Noise ● 4 Falsifiable Hypotheses
Product ResearchOpen Source

Trust Is the Product

Ideation and research behind a trust-first bounty aggregator: the polluted label space, the bounty-plaza crypto-deposit scam we reverse-engineered, and the four falsifiable hypotheses that shaped the build.

20267 min read →
Part 2 · Architecture Pure-Function Moat ● 11/11 Trust Tests
FastAPISecurity

The Moat Is a Pure Function

The async FastAPI + Next.js architecture, a natural-key data model, and a three-layer trust engine that rejects farms and scams before any network call — validated by unit tests and a live re-probe that dropped a farm's admitted count from 20 to 0.

20269 min read →
Part 3 · Lifecycle 10 open / 707 done ● No Fabricated Win
Bounty LifecycleHonest Negatives

The Honest Negative Result

The end-to-end bounty lifecycle run live: public boards on hot orgs are saturated red oceans. Why the failure created GitScout's second product axis — a viability signal — and why no completed claim was manufactured.

20268 min read →
White papers & deep-dives
White Paper · Multi-Agent 10 Autonomous Agents ● Newsroom Loop
Multi-AgentLLM OrchestrationFallback

Ten Agents, One Newsroom: Orchestrating an Autonomous Video Factory

How ten specialised AI agents hand off work like a production company — a provider-agnostic brain with automatic fallback, prompt-injected craft playbooks, a learning loop, and a QA gate that blocks broken output.

Aug 202613 min read →
White Paper · Resilience 3-Tier Failover ● $0 Pipeline Stack
LLM FailoverSelf-HealingZero-Cost

Never Return Empty: A Zero-Cost AI Video Pipeline Engineered to Always Ship

One channel URL in, a finished narrated video out — for $0. The engineering that matters isn't any single model but the resilience layer: a three-tier failover cascade, self-healing validators, and a procedural generator that guarantees output when every API fails.

Aug 202612 min read →
White Paper · Policy 230-Yr Constraint ● Divisible Power
HistoryDraft PowerGroundwater

The 230-Year Constraint: Why India's Mechanization Problem Was Never About Machines

A history-to-policy argument from 18th-century primary sources: India's mechanization bottleneck has always been access to divisible draft power — not tools, cost, or awareness — and it still is.

Aug 202616 min read →
White Paper · Full-Stack Single-Origin SSR ● EXIF Stripped · ~$3/mo
Full-StackSSRSecurity

Single Origin: A Production Civic-Reporting Platform on a Free-Tier Budget

Making a JavaScript map describe itself to WhatsApp crawlers, trusting strangers just enough to let them post, and hardening an upload path that assumes every phone photo carries home GPS — for ~$3/month.

Aug 202612 min read →
White Paper · Speech ML Demucs + IndicF5 ● Kaggle T4 Pipeline
IsochronyGemini CoTIndicF5

The Deployed Dubbing Pipeline: Fitting Translated Speech to the Frame

The production v2 system — how it constrains Gemini translation to a phoneme budget, preserves the soundstage with source separation, and clones voices at a controlled duration, all on a free Kaggle T4.

202611 min read →
Who I am

About Me

Nirmit K. Tripathii

I'm a certified AI/ML Engineer with a M.Tech in Modeling & Simulation and 3+ years of industry experience building intelligent systems across NLP, LLMs, multimodal ML, and agentic AI.

My work spans production deployments at Chetu India and BLU PARROT — including an agentic event-commerce chatbot, a Video-to-SOP multimodal pipeline, legal AI for Indian law, sports action anticipation research, and a participant in the SoccerNet 2026 Challenge.

Most recently AI/ML Software Engineer at Chetu India (through mid-2026). Now available for freelance, consulting, and research collaboration — remote — and open to full-time roles from September 2026.

  • Machine LearningRandom Forests, XGBoost, SVM, Clustering, Regression, Classification
  • NLP & LLMsRAG variants, LangChain, LangGraph, Fine-tuning, ASR, NMT, Summarization
  • Deep Learning & VisionCNNs, Transformers, Video Understanding, ImageBind, FUTR, T-DEED
  • Agentic AILangGraph StateGraph, Multi-node agents, Tool use, LangSmith observability
  • AI/ML Software EngineerChetu India · Apr 2025 – Jun 2026
  • AI/ML EngineerBLU PARROT · Jan 2024 – Apr 2025
  • Subject Matter Expert: AI/MLHouse of Coutons · Nov 2023 – Jan 2024
  • AI/ML ProgrammerHRBOTICS · Apr 2022 – Mar 2023
  • Football Data AnalystBoomer 11 · Nov 2021 – Mar 2022
  • M.Tech — Modeling & SimulationSavitribai Phule Pune University · 2019–2022
  • B.Tech — Electrical EngineeringPandit Deendayal Petroleum University · 2014–2018
What I do

Core Competencies

From classical ML to cutting-edge multimodal AI — built and shipped in production.

Agentic AI & LLMs

LangGraph multi-node agents, RAG variants (Naive/LightRAG/CRAG/VideoRAG), LangSmith tracing, OpenAI & Gemini APIs, FastAPI.

Deep Learning & Vision

CNNs, Transformers, VideoRAG, ImageBind multimodal embeddings, OpenCV, FUTR anticipation models, T-DEED temporal spotting.

NLP & Speech

BERT variants (InLegalBERT, XLM-RoBERTa, IndicBERT), FasterWhisper ASR, multilingual NMT, text classification, summarization.

Data & MLOps

FAISS, HNSWLib, Neo4j, NetworkX, HuggingFace, W&B, Python scraping, Azure OCR, data engineering pipelines.

Sports AI

Expected Goals (xG) modelling (89.77% accuracy), FAANTRA ball action anticipation, SoccerNet 2026 Challenge participant, pass network analysis.

Legal AI

LegalGPT: case summarization, 150+ contract types classification, compliance AI, Opennyai, InCaseLawBERT fine-tuning.

Where I've worked

Experience

Building AI that ships — from startup chaos to enterprise scale.

Chetu IndiaApr 2025 – Jun 2026
AI/ML Software Engineer
Built DIGEST — agentic LangGraph chatbot for event discovery & ticket commerce with full Automatiq B2B transaction lifecycle.
Architected SOP CHATBOT — multimodal VideoRAG pipeline converting screen-recorded workflows to structured SOPs via ImageBind, Gemini & LangSmith.
Built IKS Oracle (Gyanstrot) — RAG-powered AI system for Indian Knowledge Systems, enabling semantic querying of Vedic & ancient Indian texts.
BLU PARROTJan 2024 – Apr 2025
AI/ML Engineer
LegalGPT — full RAG pipeline for Indian legal research, contract analysis (150+ types), and compliance checking.
Hindi ASR portal with SoTA Conformer-CTC + Gemini NLP call summarization.
22-language Indic translation portal (IndicBERT + Indic-TransV2).
Promoted to Project Lead for UK real-estate multimodal regression model (Inferred Theorem).
HRBOTICSApr 2022 – Mar 2023
AI/ML Programmer
Multi-threaded Python scrapers for NCR commercial data and automated LinkedIn hiring workflows via Selenium + BeautifulSoup.
BOOMER 11Nov 2021 – Mar 2022
Football Data Analyst
Expected Goals (xG) model with 89.77% accuracy using RF Classifier — published research paper.
Let's talk

Contact Me

Open to new opportunities

nirmitktripathii@yahoo.com
Delhi NCR, India · Remote
Open to freelance, consulting & research collaboration